Craniofacial ontogeny in Tyrannosauridae (Dinosauria, Coelurosauria)

ABSTRACT A study of ontogenetic variation is used to clarify aspects of tyrannosaurid taxonomy and investigate the supposed phenomenon of dwarfism in the clade. A hypothetical ontogenetic trajectory is described for the relatively well-represented taxon Albertosaurus libratus. The type specimen of the purported “pygmy” tyrannosaurid Nanotyrannus lancensis was compared with specimens of A. libratus and found to share many morphological characters that exemplify immature specimens of the latter taxon. Most of the cortical surface of the Cleveland skull of N. lancensis has immature bone grain. Also, the skull shares unique derived characters with mature specimens of Tyrannosaurus rex, suggesting that the specimen is a young T. rex and not a dwarf tyrannosaurid. An increase in tooth width, accompanied by loss of tooth positions, and a global shift from an immature gracile to a mature robust morphotype in the craniofacial skeleton typifies the ontogenetic changes in T. rex. Similarly, on the basis of immature ...


INTRODUCTION
Since the turn of the century, articulated tyrannosaurid fossils have been recovered from Late Cretaceous sediments from western North America and central Asia. In particular, the Hell Creek (and equivalents) and Dinosaur Park formations of North America, and the Nemegt Formation of Mongolia, have yielded the greatest quantity of specimens including partial growth se ries for some taxa (Rozhdestvensky, 1965;Russell, 1970;Ma leev, 1974).
In 1923, Matthew and Brown were the first to hypothesize immature features in a tyrannosaurid, the type specimen of Gorgosaurus sternbergi Matthew and Brown, 1922 (AMNH 5664), noting the lightly built jaws, slender muzzle, long and low max illa, and round orbit. Despite its similarity to known specimens of G. libratus Lambe, 1914, the taxon was retained on the basis of its small size and slender proportions (Matthew and Brown, 1923). Rozhdestvensky (1965) proposed that the holotype speci mens of Tarbosaurus efremovi Maleev, 1955a, Gorgosaurus lancinator Maleev, 1955a, and Gorgosaurus novojilovi Maleev, 1955a, were ontogenetic variants of the same taxon, Tarbosau rus bataar Rozhdestvensky, 1965. He noted that young tyrannosaurids had slender jaws, an elongate tibia and metatarsal bundle, weakly developed limb joints, an elongate humerus, a columnar metatarsal III, and slightly different cranial propor tions relative to older specimens. On the basis of this work, he suggested that Gorgosaurus lancensis Gilmore, 1946, is a young individual of Tyrannosaurus rex Osborn, 1905. In 1970, Russell published a comprehensive review of tyr annosaurid taxa of western Canada. On the basis of the simi larity of Gorgosaurus to Albertosaurus Osborn, 1905, the for mer genus was considered a subjective junior synonym of the latter. The holotype of G. sternbergi was recognized as a ju venile and referred to A. libratus (Russell, 1970). Russell iden tified limb proportions, degree of ossification of the pelvic bones and the skull roof, allometric changes in the skeleton, and the development of the nuchal crest and jugal cornual pro cess as indicators of relative maturity among tyrannosaurids.
Based on a growth series of T. bataar, Carpenter (1992) ob served that during ontogeny tyrannosaurid skulls deepen, the muzzle shortens, the orbit becomes dorsoventrally elongate, the supraorbital region of the skull becomes progressively rugose, the metatarsus shortens, and the neurocentral and calcaneoas-tragalar sutures fuse with age. Like Rozhdestvensky (1965), he implicated ontogenetic variation in Maleev's splitting of T. ba taar into several taxa.
Despite these treatments of tyrannosaurid ontogeny and sub sequent revisions of alpha taxonomy, consensus over the latter among workers is elusive (Bakker et al., 1988;Paul, 1988;Car penter, 1992;Makovicky and Currie, 1998). Also, no study has documented the bone-by-bone changes in the tyrannosaurid cra niofacial skeleton despite the wealth of material from the Di nosaur Park and Nemegt Formations. Documentation of sizeindependent morphological changes may facilitate the recog nition of diagnostic and phylogenetically informative charac ters, thereby contributing to the resolution of the current lack of consensus over tyrannosaurid taxonomy.
The purposes of this paper are: (1) to describe the ontoge netic changes in the skull bones of the well-represented taxon Albertosaurus libratus, (2) to describe taxonomic differences between taxa; and (3) to test the hypothesis of dwarf tyranno saurids. Institutional

Materials
This study is based on observations and measurements of 46 tyrannosaurid specimens housed in several North American in stitutions (Table 1); all other specimens noted were observed from published photographs and figures. Tyrannosaurid cranial remains from the Dinosaur Park and Nemegt formations present a range from small (skull length 400 mm) to large (skull length 1,050+ mm) specimens. Skulls recovered from the Hell Creek Formation and equivalents are usually over 1 meter long, and only one small skull (skull length 572 mm) has been recovered to date.

Methods
The present analysis is descriptive, with an emphasis on the hypothetical ontogenetic changes to discrete morphological fea tures. For comparison, teeth were measured mesiodistally and labiolingually at the crown base, immediately proximal to the distal carina. Terminology is after Bakker et al. (1988), Baumel and Witmer (1993), and Witmer (1997).
Following Bennett (1993) and Sampson (1993), cortical bone texture was used as a size-independent criterion of maturity where possible. Young ornithodirans are characterized by stri ated cortical bone that follows the direction of growth. Young tyrannosaurids also display striated cortical texture on virtually every bone. However, on many display specimens observation of bone texture was difficult or impossible to see given the distance imposed by glass cases and layers of consolidant or paint. Thus, a rigorous documentation of bone grain transition like that done for ceratopsids by Sampson (1993) could not be obtained.
Tyrannosaurids are unlike ceratopsids (Sampson, 1993) in their patterns of sutural fusion. In large tyrannosaurid speci mens, fusion between endochondral (e.g., prootic, supraoccipital, otoccipital) occipital bones occurs, but without an evident pattern. Otherwise, the sutures between dermal and endochon dral bones remain open. Sutural fusion may be an informative independent criterion for assessing relative maturity in tyran nosaurids, but in many of the specimens examined, sutures be tween braincase and occipital bones were obscured by glass cases, incomplete preparation, plaster restoration, or layers of consolidant.
In the present study, the hierarchically nested, progressive development of size-independent ontogenetic characters was used to identify ontogenetic stages. One young specimen, ROM 1247, was chosen for its completeness and immaturity (indi cated by the presence of immature bone grain on each bone) for comparison with other A. libratus specimens. The specimen was described in detail as the reference for comparison with all other specimens. Features of other specimens that were less developed than those of ROM 1247 were considered immature; features showing greater development were considered more mature. For bones that were missing in ROM 1247 (e.g., post orbital, palatine, squamosal), specimens of the same ontoge netic stage (e.g., AMNH 5664) were used for comparison with other specimens for those bones.
On the basis of the changes of discrete morphological char acters, the ontogeny of A. libratus was divided into three stages. Stage 1 specimens have (where observable) striated cortical bone grain and express nascent ontogenetic characters. One ex ceedingly small specimen (TMP 94.12.155), representing a skull of ~370 mm in length and consisting of a pair of man dibular rami, displayed features less developed than in the larg er Stage 1 specimens. Since the entire skull was not represent ed, and only one informative character was noted, the specimen was not allocated to its own growth stage; ROM 1247 itself was regarded as a large Stage 1 specimen.
Stage 2 specimens exhibit development of the homologous features that typified Stage 1. Stage 2 specimens may also retain Stage 1 features, displaying a transitional constellation of char acters between Stages 1 and 3. Stage 3 specimens express fur ther development of the features that typified Stage 2 and dis play development of or retain Stage 1 features. Finally, Stage 4 is represented exclusively by adult Daspletosaurus and Ty rannosaurus specimens that display homologous ontogenetic features that are more developed than those of Stage 3 A. li bratus specimens. The ontogenetic stages are of some degree arbitrary, but a quantitative analysis of the ontogenetic char acters following Brochu (1996) is forthcoming to determine the relative maturity between individual specimens.

TAXONOMY OF DINOSAUR PARK AND HORSESHOE CANYON TYRANNOSAURIDAE
The holotype of A. libratus (CMN 2120) was first named Gorgosaurus libratus by Lambe (1914). Its slender upper incisiform teeth distinguished it from the robust dentition of Deinodon horridus Leidy, 1856 from the equivalent Judith River Formation of Montana. Later, Matthew and Brown (1922) showed that the type teeth of Deinodon and Gorgosaurus were indistinguishable, and that the two taxa likely were congeneric. However, the authors retained generic distinction in the absence of diagnostic skeletal material for Deinodon. Russell (1970), in agreement with Gilmore (1946), determined that the lectotype of D. horridus, comprising two incomplete incisiform teeth, is indistinguishable from the Dinosaur Park material, and D. hor ridus is therefore a nomen dubium (for an opposing view, see Sahni, 1972). Russell (1970) concluded that Gorgosaurus libratus and Albertosaurus sarcophagus Osborn, 1905 were congeneric on the basis of overall similarity, a position that is upheld herein on a phylogenetic systematic basis (see Carr, 1996). Albertosaurus was rediagnosed and distinguished from Daspletosaurus by Russell (1970).
I agree with Russell (1970) that the types and referred ma terial of G. libratus and A. sarcophagus appear to be identical, and are different from specimens referred to Tyrannosaurus bataar Maleev, 1955b andT. rex Osborn, 1905. I also agree with Carpenter (1992) that discrete differences exist between the tyrannosaurid skulls from the Dinosaur Park and Horseshoe Can yon formations.
Ontogenetic changes to discrete morphological characters are based on specimens referred to A. libratus, with the exception of FMNH PR308. Morphologically, this specimen is identical to the morphotype represented by D. torosus. Although the skull is considered representative of A. libratus (Russell, 1970: fig. 1;Paul, 1988:335;Carpenter, 1992:figs. 1, 2E), the speci men is less complete (Fig. 1A, B) than usually shown. Also, FMNH PR308 has been focal in discussions of tyrannosaurid diversity, with specific reference to tooth size and number (Bakker et al., 1988;Paul, 1988). In fact, every upper tooth and all but 13 dentary teeth are restored in plaster (pers. obs.), render ing the material basis of the former aspect moot.
The relationship between size and morphology has not been adequately studied in large theropods, and no attempt has been made to do so herein. This important question would best be answered by a quantitative and comparative study in the realms of biomechanics and functional morphology. The ontogenetic characters in this study were chosen cognizant of this issue. Thus, inclusive features of possible structural importance to the skull were avoided.

TAXONOMY OF NEMEGT TYRANNOSAURIDAE
I agree with Paul (1988) and Carpenter (1992) that Tarbosaurus is a junior subjective synonym of Tyrannosaurus (contra Rozhdestvensky, 1965). Both species are united by the absence of a lacrimal horn (Carpenter, 1992), inflated descending paroccipital process, transversely-oriented occipital plate, and rostrocaudally restricted basisphenoid recess, among other fea tures. A more detailed treatment of tyrannosaurid systematics is forthcoming.

Ontogenetic Variation in Albertosaurus libratus
Premaxilla-The premaxilla of Stage 1 specimens of A. li bratus is transversely narrow with a concave lateral margin, has a shallow alveolar process (Fig. 5A), narrow maxillary process (Fig. 5C), and the maxillary articular surface of the alveolar process is transversely narrow. The premaxillae of Stage 2 spec imens are not distinctive. In Stage 3 specimens, the premaxillae are transversely broad in rostral view, which straightens the lateral margin of the bone.
Maxilla-In small Stage 1 specimens (e.g., CMN 12063) of A. libratus, the maxilla is transversely compressed. In large Stage 1 specimens (e.g., ROM 1247) in lateral view, the bone is thickened, and the slot for the maxillary process of the nasal is dorsolateral in position (1; Fig. 2E). In small Stage 1 speci mens, the margin of the antorbital fossa is sharply delimited, and its ventral margin may pass caudally in a concave arc, or is straight; in larger specimens, the rostroventral margin of the fossa grades into the textured bone surface (2; Figs. 2E, 5A). In small specimens, the base of the interfenestral strut is flat, but is gently concave in large specimens (3; Figs. 2E, 5A).
In Stage 1 specimens, the antorbital fenestra is longer than high (Fig. 5A). The lateral surface is textured and incised by shallow neurovascular sulci, the ventral margin of the antorbital fossa is bounded by a low ridge (4), and the alveolar process is shallow (5; Figs. 2E, 5A). The vestibular bulla and passage of the subnarial foramen are laterally flat or transversely convex (6; Fig. 2E).
Also in Stage 1 specimens, the rostroventral foramen of the premaxillary process is small and the rostrodorsal foramen is a cleft-like, ventrally opening slit (Fig. 5A). The caudal foramen of the ventral row of foramina bears a caudal elongate sulcus, that does not leave the ventrolateral margin of the jugal process (7; Figs. 2E, 5A). The maxillary fenestra is positioned midway between the rostral margins of the antorbital fenestra and fossa, and is as long as deep or barely longer than high (8; Figs. 2E, 5A). The lateral surface separating the antorbital fossa and nasal suture is a shallow tab (9; Figs. 2E, 5A). Finally, the promaxillary fenestra is a slit-like foramen within the rostral margin of the antorbital fossa (10; Figs. 2E, 5A) (Russell, 1970).
In Stage 2 specimens (e.g., AMNH 5336), the maxilla is thickened laterally and bowed rostrally (11; Fig. 2F). The rostrolateral surface is expanded rostrally and dorsoventrally (Fig. 2F) and its surface sculpturing is pronounced. The ventral mar gin of the antorbital fossa is either gently sigmoid or dorsally convex (Fig. 2F). The depression ventral to the antorbital fossa is shallow (12; Fig. 2F). The maxillary fenestra is longer than high and approaches the rostral margin of the antorbital fenestra (13; Fig. 2F). The promaxillary fenestra is recessed dorsally (14), and the lateral surface of the maxilla passes over the ros tral margin of the antorbital fossa as a strut (15; Fig. 2F). The rostroventral foramen of the premaxillary process is larger than the round rostrodorsal foramen. The ventral rim of the ventral jugal process is breached by the sulcus from the caudal foramen of the ventral row of foramina (16; Fig. 2F).
In Stage 3 specimens (e.g., AMNH 5458), the rostrolateral surface of the bone is thickened, bowed, and expanded, dis placing the articular surface for the nasal dorsally. The rostro lateral foramina are enlarged. The base of the interfenestral strut is concave, and the height of the antorbital fenestra approaches its length.
In medial view in A. libratus, shallow pneumatic excavations are present in the maxillary antrum in Stage 1 specimens (e.g., ROM 1247; Fig. 21). The caudoventral excavation may be shal low or pocket-like (17; Fig. 21). The floor of the promaxillary antrum is crossed by a low ridge above the third alveolus. The palatal process has a strong rostrodorsal sigmoid curvature (18); its caudal surface is flat such that its dorsal surface is visible (19) and its ventral margin extends beneath the level of the medial alveolar process (20; Figs. 21, J). The medial edge of the palatal process is cleaved by the articular surface for the palatine (21; Figs. 21, J). Also, the palatal process is transverse ly narrow; the tooth root bulges are low (22; Fig. 2J), and the interdental depressions are shallow.
In Stage 2 specimens (e.g., ROM 4591), the nasal is thick ened rostrally and the ventral surface is not strongly vaulted. The caudal plate expands between the lacrimals. The medial frontal process may be absent in Stage 2 specimens (e.g., AMNH 5336). In Stage 3 specimens, the lateral margins of the caudal plate expand or are parallel (e.g., CMN 2120) between the lacrimals.
In small Stage 1 specimens, the rostral ramus is divided into lateral and medial processes; the former is situated dorsal to the latter, so that the ramus is forked in lateral view (31; Fig. 3A). In larger Stage 1 specimens, the processes overlap in lateral view, losing the forked appearance (32; Fig. 3B). In small spec imens, the lacrimal antorbital fossa of the dorsal ramus is fully exposed to view (33; Fig. 3A). In larger specimens, the lateral external surface is extruded ventrally as a lamina, partly con cealing the fossa in lateral view, creating a slot-like passage (31; Figs. 3B, 5A).
In Stage 1 specimens, the textured dermal surface and the smooth fossa surface merge at their juncture (32; Fig. 2A). In Stage 1 specimens, the rostral margin of the ventral ramus merges with the ventral lip of the lacrimal antorbital fossa (33; Fig. 3A, B, 5A). Also, the rostral margin of the rostroventral lamina of the lacrimal is straight or concave in lateral view (34; Figs. 3A, B, 5A). The jugal articular surface of the rostroventral lamina exceeds that of the ventral ramus. The caudal margin of the jugal articular surface of the ventral ramus is sub vertical (Fig. 5A).
In Stage 2 specimens (e.g., AMNH 5336), the ventral process of the rostral ramus is developed, but is shorter than the dorsal process (35; Fig. 3C). The cornual process may be bulbous and has one apex (36; Fig. 3C). The ventral margin of the lateral lamina of the rostral ramus matches that of the lacrimal antor bital fossa (37; Fig. 3D). The caudal margin of the jugal artic ular surface of the ventral ramus slopes caudodorsally (38; Fig.  3E). This is also in D. torosus (39; Fig. 3H) The rostral margin of the ventral ramus is embayed by the lacrimal antorbital fe nestra (40; Fig. 3C).
In Stage 3 specimens (e.g., CMN 2120), both the dorsal and ventral processes of the rostral ramus are elongate (41; Fig. 3E). The cornual process bears a single erect apex, situated above the ventral ramus (42; Fig. 3E). The lateral surface around the lacrimal pneumatic recess is not dished (43; Fig. 3E). The lac rimal antorbital fossa is attenuated rostrally by the lateral lam ina (44; Fig. 3E). This condition is also seen in Stage 4 D. torosus (45; Fig. 3G) and T. rex. In A. libratus, it is equivocal whether or not the lateral and medial laminae are fused ven trally; the specimen in which this is observed (CMN 2120) is mediolaterally crushed in this region (Fig. 3E).
In Stage 3 specimens (e.g., AMNH 5458), an edge separates the rostral margin of the ventral ramus from the lacrimal an torbital fossa. The rostral margin of the rostroventral lamina is convex in lateral view (46; Fig 3E) and the extent of its contact with the jugal is matched by that of the ventral ramus (47; Fig.  3E). This is also in Stage 4 D. torosus (Fig. 3F).
The caudal margin of the lacrimal articular surface is sub vertical in lateral view (52; Figs. 3I, 5A). The articular surface for the postorbital is shallow and extends to the ventral orbital margin (53; Figs. 3I, 5A). The medial articular surface for the lacrimal is overlapping. The caudal margin of the postorbital ramus is sinuous or concave (54; Figs. 3I, 5A). The area ventral to the postorbital ramus is flat or convex in lateral view (55; Figs. 3I, 5A). The transversely flat cornual process may be pro nounced or its caudal margin may be low (56; Figs. 3I, 5A). Finally, the quadratojugal articular surface passes rostrodorsally at or caudal to the midlength of the ventral process of the quad ratojugal ramus, either horizontally or at a steep angle (57; Figs. 3I, 5A).
In Stage 3 specimens, the maxillary ramus is dorsoventrally deep (61; Fig. 3K). The caudal margin of the jugal foramen is resorbed, exposing the rostral margin of the secondary fossa (62; Fig. 3K). The lateral surface at the base of the postorbital ramus is shallowly concave (63; Fig. 3K). The caudodorsal margin of the jugal pneumatic recess merges with the lateral surface of the jugal beneath the lacrimal contact (64; Fig. 3K). Finally, the caudal margin of the lacrimal articular surface slopes caudodorsally along an elongate and shallow lateral overlap (65; Fig. 3K).
Postorbital-In lateral view, the postorbital of small Stage 1 specimens of A. libratus is a slender and delicate bone (Fig.  3M). The frontal ramus approaches the length of the squamosal and jugal rami (66; Fig. 3M). The squamosal ramus is slender and arched in lateral view (67; Fig. 3M). The cornual process is a low, striated depression at the caudodorsal margin of the orbit (68; Fig. 3M). The jugal ramus is elongate and slender (69; Fig. 3M), and the jugal articular surface is shallow. The lateral surface of the jugal ramus bears dorsally arched sulci (70; Fig. 3M). The rostral and caudal margins of the jugal pro cess are parallel in lateral view (71; Fig. 3M) and reach the ventral margin of the orbit. In dorsal view, the dorsotemporal fossa is shallow and is not bounded rostrally by a ridge.
In large Stage 1 specimens, the ventral margin of the squa mosal process is sinuous (72; Fig. 3N). In Stage 1 specimens, the dorsal margin of the bone tends to be vertically oriented ( Fig. 5C). In large Stage 1 specimens, the incipient cornual process is a flattened, ear-like tab of bone with a horizontal ridge beneath its dorsal margin (75; Fig. 3N). The dorsallyarched sulci reach its rostral and caudal margins (74; Fig. 2P). The squamosal articular surface extends forward of the rostral margin of the laterotemporal fenestra. The frontal ramus is short in lateral view (76; Fig. 3N), and is broad in dorsal view with a moderately deep dorsotemporal fossa bounded by a low ridge rostrally.
In Stage 2 specimens, the sulci of the jugal ramus are re stricted rostral to its caudal margin (77; Fig. 30). The laterodorsal bone margin is everted medially (78; Fig. 30) and the bone ends above the orbit floor. The frontal ramus is stout and deep in lateral view (79; Fig. 30). The cornual process is prom inent (80). In Stage 3 specimens, the cornual process of the postorbital may be enlarged, consisting of a thick ridge sepa rated by a deep crease from an enlarged caudoventral boss. This is also true for D. torosus (81; Fig. 3P).
Frontal-In large Stage 1 specimens (e.g., ROM 1247) of A. libratus, the orbital margin is within a vertical slot between the articular surfaces for the lacrimal and postorbital in lateral view (Figs. 5A, C). The lacrimal notch is long and narrow in dorsal view (Fig. 5C). The paired frontals are longer than wide (Fig. 5C). In Stage 1 specimens, the dorsotemporal fossa is shallow (Fig. 5C). In Stage 2 specimens (e.g., AMNH 5336), the fossa is deep. In Stage 1 specimens, the frontals are flattened to meet at the midline (Fig. 5C). In Stage 2 specimens, the frontals slope dorsomedially to their contact.
In Stage 1 specimens (e.g., ROM 1247) of A. libratus, the prefrontal articular surface is narrow in dorsal view.
Parietal-In Stage 1 specimens (e.g., ROM 1247) of A. li bratus, the nuchal crest is low in caudal view, only as deep as the dorsal process of the supraoccipital. The rostrodorsal sur face of the crest is rugose laterally and the dorsolateral margin of the crest flares caudolaterally (Fig. 5C). The sagittal crest is low (Fig. 5A).
In Stage 2 specimens (e.g., AMNH 5336), the rostrodorsal margin of the nuchal crest is rugose to the midline and the dorsolateral margin is rugose caudally. The dorsal margin of the sagittal crest is concave in lateral view. In Stage 3 speci mens (e.g., AMNH 5458), the nuchal crest is tall, twice as deep as the dorsal process of the supraoccipital.
Basioccipital-In Stage 1 specimens (e.g., ROM 1247) of A. libratus, the occipital condyle is caudoventrally flattened and the lateral margins converge ventrally. The ventral surface of the basituberal web is flat and arches dorsally in caudal view. The surface beneath the occipital condyle is convex in frontal section and flares ventrally between the ascending scars (sensu Bakker et al., 1988). Finally, the basal tuber is poorly devel oped.
In Stage 2 specimens (e.g., AMNH 5336), the occipital con dyle is spherical. The basal tuber forms a rugose block. In Stage 3 specimens (e.g., AMNH 5458), the caudal surface of the ba sioccipital is concave ventral to the occipital condyle.
Basisphenoid-In Stage 1 specimens (e.g., ROM 1247) of A. libratus, the basipterygoid process is flattened rostrolaterally, the basisphenoid pneumatic foramina are small and set above the ventral margin of the basipterygoid web. In lateral view, the ventral margin of the bone slopes at a low rostroventral angle such that the dorsal margin of the basipterygoid process does not project below the level of the caudo ventral comer of the bone. The oval scar (sensu Bakker et al., 1988) is smooth, ventromedially oriented, and narrow.
In Stage 2 specimens (e.g., AMNH 5336), the oval scar is broad. In Stage 3 specimens (e.g., AMNH 5458), the ventral margin of the basisphenoid descends steeply rostroventrally in lateral view such that the dorsal margin of the basipterygoid process extends ventral to the level of the caudoventral corner of the bone. The oval scar is laterally expanded, ventrally ori ented, and dished.
Vomer-In lateral view, the vomer of Stage 1 specimens (e.g., ROM 1247) of A. libratus has a horizontal ventral margin which curves gently caudoventrally, caudal to midlength (83; Fig. 4A). A slender neck is present between the transverselyexpanded maxillary process and the dorsally deep body of the bone (82; Fig. 4A). In Stage 2 specimens (e.g., AMNH 5336), the ventral margin curves strongly caudoventrally behind the midlength of the bone.
Palatine-In small Stage 1 specimens (e.g., TMP 86.144.1) of A. libratus, the caudalmost of the two pneumatic recesses in the lateral surface of the bone is rostrocaudally elongate and vertical struts are present on the medial wall of the rostral re cess. The septum between the pneumatic recesses is narrow. Also, the palatine body is transversely compressed. In larger Stage 1 specimens (e.g., USNM 12814), the palatine is trans versely inflated. In small Stage 1 specimens, the vomerine ra mus is dorsoventrally shallow in lateral view; it is deep in larger Stage 1 specimens.
Surangular-In small Stage 1 specimens (e.g., TMP 86.144.1) of A. libratus, there is no ridge lateroventral to the glenoid in lateral view and the caudal margin of the retroarticular process is concave. In small and larger Stage 1 specimens (e.g., ROM 1247), the bone is shallow, such that the rostroven tral margin is convex and slopes at a low angle caudoventrally (84; Figs. 4E, 5B). The intramandibular process is stout and deep, its ventral margin meets the rostroventral margin of the bone along a shallow curve, or is confluent (85; Figs. 4E, 5B). The rostral plate is externally flat (86; Figs. 4E, 5B). The sur angular shelf slants rostroventrally (87; Figs. 4E, 5B) or hori zontally; its lateral margin projects laterodorsally. The dorsomedial flange may be low and blade-like (88) or high (TMP 86.144.1) with a narrow shelf separating it from the surangular shelf, in which the insertion scar of M. adductor mandibulae extemus (Molnar, 1991) is indistinct and confined medial to the surangular shelf (Figs. 4E, 5B).
In Stage 2 specimens (e.g., AMNH 5336), the bone is deep such that the rostroventral margin slopes at a steep angle cau doventrally. The ventral margin of the intramandibular process and rostroventral margin of the surangular meet at an angle.
The surangular shelf passes horizontally, rostroventrally, or rostrodorsally, and its lateral margin projects horizontally. The sur angular foramen is large and recessed. The sulcus lateroventral to the glenoid is dorsoventrally deep and rugose.
In Stage 3 specimens (e.g., CMN 2120), the intramandibular process is elongate and meets the rostroventral margin at an angle (91; Fig. 4F). The surangular shelf passes rostrodorsally (92; Fig. 4F) and the dorsomedial process is tall. The dorsolat eral muscle scar is delimited rostrally by a sharply inset facet, which extends to the lateral surface of the surangular shelf (93; Fig. 4F). The caudal margin of the retroarticular process is con cave.
Prearticular-In small Stage 1 specimens (e.g., TMP 86.144.1) of A. libratus, the ventral portion of the articular sur face of the angular is reduced and the lateral surface of the contact is aliform. In Stage 1 specimens (e.g., ROM 1247), the dorsal margin of the caudal ramus is restricted caudally (94; Fig. 4J). The caudal ramus is shallow with parallel dorsal and ventral margins (95; Fig. 4J). The rostral lamina is strap-like (96) and pointed distally (97); its caudodorsal margin is smooth (Fig. 4J). In large Stage 1 specimens, the angular facet is flat and not aliform.
In Stage 3 specimens (e.g., CMN 2120), the dorsal margin of the caudal ramus, with the adductor attachment surface, is shifted rostrally toward the mid-shaft (98; Fig. 4K). The caudal ramus is deep such that the dorsal and ventral margins are gent ly convex and converge rostrally (99; Fig. 4K). The caudodorsal margin of the rostral lamina bears a rugose surface for muscle attachment.
Dentary-In small Stage 1 specimens (TMP 94.12.155) of A. libratus, the shallow dentary is as wide as it is deep. In larger Stage 1 specimens (e.g., ROM 1247), the dentary is deeper than wide. The angular process is dorsoventrally shallow (Fig. 5B). The symphyseal facet is flat and textured by stout caudodorsal bony papillae. The splenial articular surface is indicated by light, arcuate rostroventral ridges. The ventral bar beneath the Meckelian foramen is lightly rugose and rostroventrally exca vated by a broad sulcus. Usually in tyrannosaurids the angular process is bifurcated by the external mandibular fenestra in lat eral view; in ROM 1247 (Fig. 5B) this emargination is absent on both sides. In Stage 2 specimens (e.g., AMNH 5336), the angular process is deep. In Stage 3 specimens (e.g., AMNH 5458), the symphyseal surface may be rugose.

Taxonomic Variation
Premaxilla-In rostral view, the premaxillae of the holotype (Stage 4) of D. torosus (CMN 8605) are fused by struts of bone at the base of the nasal process; however, it is possible that the trabeculae are pathological exostoses. In addition, "skirts" of bone extrude from the external margin of the alveoli, and the alveolar process is deep (see Appendix 1 for a synoptic com parison of taxonomic variation).
In caudal view ( Fig. 2A), the medial process of the premax illa of D. torosus has a pronounced dorsal flange (100), and there is a crest-like ridge that delimits the subnarial foramen ventrally (101; Fig. 2A). In D. torosus and Stage 4 T. rex spec imens, the maxillary process is broad, elongate, and flattened ( Fig. 8C, I). In T. rex the maxillary process broadly overlaps the medial edge of the maxillary process of the nasal.
Maxilla----In lateral view in Stage 4 specimens of D. torosus (e.g., CMN 8506), the maxilla is thickened transversely, elim inating the depression ventral to the antorbital fossa, and the ventral floor of the fossa is ledge-like (102; Fig. 2G). The lateral surface sculpture is pronounced, with deep neurovascular sulci (103; Fig. 2G). The nasal articular surface is displaced medially (104; Fig. 2G). The vestibular bulla is swollen and convex in transverse section (105; Fig. 2G) and its foramina are separated by the rugose lateral surface. The lateral surface separating the antorbital fossa from the nasal suture is as deep as the fossa beneath it (106; Fig. 2H). The articular surface for the nasal may be transversely broad, forming a deep peg-and-socket ar ticulation (107; Fig. 2G, K).
The alveolar process of the maxilla in D. torosus is deep and expanded; the dorsal and ventral margins gradually con verge caudally (108; Fig. 2G). Alveolar skirts are pronounced (109; Fig. 2G). The first tooth is subincisiform and the suc ceeding teeth have subconical crowns. In contrast, the first tooth in A. libratus is incisiform (110; Fig. 2E), and the crowns of successive teeth are labiolingually compressed. The ventral margin of the antorbital fossa passes caudally along a dorsally convex arc (102; Fig. 2G). The antorbital fenestra is as tall as long (see Russell 1970:fig. 6). The maxillary fenestra is rostrally elongate and dorsoventrally deep (110), forming a strut medial to the rostral fossa margin (Fig. 2G). The promaxillary fenestra lies between these laminae and is dorsally recessed (111; Fig. 2G).
In medial view in D. torosus (e.g., CMN 8506), the dental impressions are pit-like (112; Fig. 2K). The dorsal margin of the maxillary antrum passes rostroventrally (113; Fig. 2K), whereas in A. libratus the margin is horizontal, beneath which the maxillary fenestra is situated up to half of its height below (114; Fig. 21). In D. torosus, the dorsal margin of the maxillary fenestra approaches (115; Fig. 2K) or extends to the dorsal mar gin of the antrum. The caudoventral excavation of the maxillary antrum is enlarged and deep (116; Fig. 2K). The floor of the promaxillary recess is crossed by a strut above the third alve olus (117; Fig. 2K).
The palatal process is wide and bounded by a distinct medial ridge; its ventromedial margin passes above the ventral margin of the alveolar process (118; Fig. 2K). The palatine articular surface is a dorsoventrally deep and flat facet (119; Fig. 2K). The tooth root bulges are not visible caudally on the palatal process (120; Fig 2K). The interdental plates are enlarged but unfused (121; Fig. 2K); in some specimens (e.g., AMNH 3456) only the triangular apices of the rostral plates are visible.
In Stage 4 specimens of T. rex (e.g., AMNH 5027), the cau dal plate is reduced to an elongate rod by the medially expanded lacrimals (Fig. 8C) and the articular surface for the maxilla is a peg-and-socket contact (Fig. 8E).
Lacrimal-In D. torosus (e.g., CMN 8506), the cornual pro cess is inflated dorsally and transversely (125; Fig. 3F), broad ening its caudal frontal contact, and eliminating the shelf above the ventral ramus (126; Fig. 3F). This inflated condition is also seen in the homologous region of Stage 4 T. rex specimens (Fig. 8C, E, I). In D. torosus, the cornual process is twice as deep as the lacrimal pneumatic recess (127; Fig. 3F, G). The Tshape of the bone is obscured by the inflated rostral and supra orbital rami (128; Fig. 3F). The rostral margin of the ventral ramus is embayed by the lacrimal recess (129; Fig. 3F). The lacrimal antorbital fossa and the lateral surface of the ventral ramus are sharply separated by the leading edge of the ventral ramus (129; Figs. 3F, G).
The lateral lamina of the rostral ramus appears to be fused with the medial lamina ventrally, rostral to the lacrimal pneu-  Fig. 3F, G). The dorsal process of the rostral ramus is elongate in contrast to the short ventral process (130; Fig. 3G). The ventral ramus is rostrocaudally broad beneath the dorsal ramus (Fig. 3F). The nasal articular surface is deep and over-and underlaps the bone (CMN 8506). The rostral margin of the rostroventral lamina is straight or convex (131; Fig. 3F). This condition is also in Stage 4 specimens of T. rex (Fig. 8E).
Albertosaurus libratus, A. sarcophagus, and D. torosus all possess lacrimal cornual processes, in which the cornual pro cess projects rostrodorsally, the apex set rostral or dorsal to the ventral ramus. In D. torosus, the rostral ramus is inflated such that the horn is not markedly offset in lateral view. The apex of A. sarcophagus is reduced, offset rostrally by a gentle emargination.
In both species of Tyrannosaurus, there is no cornual process (Fig. 8C, E, I). Also, the dorsal ramus of Stage 4 specimens is inflated, such that the bone is not dished above the small lac rimal pneumatic recess (Fig. 8E). In Stage 4 T. rex specimens, it appears that the medial and lateral laminae of the rostral ramus are fused, and the surface of the latter is excavated lat erally by the antorbital fossa and is pierced by two pneumatic foramina (Fig. 8E). In Albertosaurus and D. torosus, the ventral and dorsal rami meet at a right angle (Figs. 2G-K, 3G-I, 5A). In Tyrannosaurus, the ventral ramus meets the dorsal at an acute angle (Molnar, 1991) (Fig. 8E).
Jugal-In D. torosus (e.g., CMN 8506), the cleft between the processes of the maxillary ramus curves caudodorsally in lateral view. In A. libratus, the cleft is horizontal in lateral view (132; Fig. 3I). The antorbital and secondary fossae of D. to rosus are separated by a pronounced, rounded septum (133; Fig.  3L) (except TMP 85.62.1). The caudal margin of the jugal pneumatic recess is rebsorbed, fully exposing the secondary fossa in lateral view (134; Fig. 3L). This condition also pertains to T. rex (Fig. 8E). The postorbital articular surface is deep and braced ventrally by a bony shelf well dorsal to the ventral mar gin of the orbit (135; Fig. 3L). The bone is strut-like beneath the postorbital articular surface, producing a deep lateral con cavity (136; Fig. 3L); a similar but less pronounced strut and depression are present in Tyrannosaurus (e.g., AMNH 5027). The cornual process is prominent and transversely swollen (137; Fig. 3L). The quadratojugal articular surface passes ros trodorsally ahead of the midlength of the ventral process (138; Fig. 3L). The medial articular surface for the lacrimal is braced ventrally by a ridge. Finally, the maxillary ramus is deepened at the level of the jugal pneumatic recess (139; Fig. 3L) and the rostral extremity of the ramus is stout (except TMP 85.62.1).
As in Albertosaurus and D. torosus, Stage 4 Tyrannosaurus specimens bear a low cornual process (Fig. 8C, E, I). In T. rex, the contribution of the jugal to the antorbital fenestra is restrict ed between the maxilla and lacrimal (Fig. 8C, E), unlike the extensive exposure in Albertosaurus (Fig. 5A) and D. torosus (Russell, 1970: fig. 6).
Postorbital-In D. torosus (e.g., CMN 8506), the cornual process may be enlarged to reach or project beyond the dorsal margin of the bone (81; Fig. 3P). The jugal ramus tapers to a point (140). In A. libratus, the distal margin of the jugal process is angular in lateral view (69; Fig. 3M). In D. torosus, the bone terminates far above the ventral margin of the orbit. Also, a small suborbital prong is present (141; Fig. 3Q). In medial view, the articular surface for the jugal is bipartite and deeply slot-like. The dorsal articular surface for the squamosal termi nates caudal to the rostral margin of the laterotemporal fenestra.
In T. rex (e.g., AMNH 5027), a dorsal and ventral differen tiation of the patch-like cornual process is evident. A rugosity is sometimes developed from its rostral half that proceeds caudally, and may bear a skirt-like ventral ridge (Molnar, 1991). In Tyrannosaurus, the suborbital prong is pronounced (Fig. 8C, E, I).
In Stage 4 specimens of T. rex, the nasal ramus is foreshort ened and the dorsotemporal fossa is deep, producing a distinct transverse bar at its rostral margin (Fig. 8C). In some specimens (e.g., TMP 81.6.1) the rostral fossa margin passes caudolaterally in dorsal view. Also in T. rex, the paired frontals are wider than each is long (Fig. 8C). The sagittal crest is concave in lateral view and the rostral extent of the crest is cleft sagittally in dorsal view (Fig. 8C).
Prefrontal-In Stage 1 specimens (e.g., ROM 1247) of A. libratus, the prefrontal is stout, slightly longer than wide in dorsal view, and does not extend far rostral to the nasal process of the frontal in dorsal view (Fig. 5C).
In small Stage 1 and Stage 4 specimens (e.g., CMN 8506) of D. torosus, the prefrontal is several times longer than wide, reaching or extending beyond the nasal process of the frontal rostrally (Russell, 1970: fig. 7). In Stage 4 specimens, the prox imal half of the bone is flat and transversely expanded (Russell, 1970: fig. 7).
In Stage 4 T. rex specimens, the prefrontal is exposed dorsally as a sliver of bone between the frontal, nasal, and lacrimal (Fig. 8C).
Parietal-In Stage 4 D. torosus specimens (e.g., CMN 8506), the nuchal crest is tall, rostrally-everted in lateral view, and dorsolaterally expanded (Russell, 1970:figs. 6, 7). The lat eral margins are deeply concave in caudal view. In lateral view, the dorsal margin of the keel-like sagittal crest is deeply con cave rostrally. The lateral contact with the frontal is strength ened by a strong transverse ridge (Russell, 1970: fig. 7).
In Stage 4 T. rex specimens, the nuchal crest may be rostro caudally thick and bears a flat and rugose dorsal platform (Fig.  8C, E, G, I). The sagittal crest is transversely thick (Fig. 7C).
Basioccipital-In Stage 4 D. torosus (e.g., CMN 8506), the basioccipital is broadly exposed on the floor of the foramen magnum between the exoccipitals. The ventral surface of the basituberal web is blade-like, as in T. rex. The basituberal web may be horizontal in caudal view. The basioccipital is exca vated by a deep pit ventral to the neck of the occipital condyle. The basal tuber may be elaborated into a rugose block.  In Stage 4 T. rex (e.g., AMNH 5027), the subcondylar recess is shallow (Fig. 8G), induced by the transversely broad and inflated paroccipital region. The recess is occupied by the small, oval basioccipital pneumatic foramen, that pierces the dorso lateral comer of the bone next to the basioccipital-otoccipital suture (Fig. 8G).
Unlike the arcuate suture of other tyrannosaurids, the otoccipital-basioccipital suture of the occipital condyle of the topotype and a referred specimen (ROM 12790) of A. sarcophagus is angular in caudal view. The basioccipital separates the otoccipitals from the midline of the condyle by a triangular process. However, the holotype of A. sarcophagus displays the condition typical of other tyrannosaurids.
Basisphenoid-In Stage 4 D. torosus (e.g., CMN 8506), the rostral margin of the cultriform process curls medially to en close the space between itself and the rostral surface of the bone. The ventral margin of the bone descends steeply rostro ventrally. The basipterygoid facet faces rostrally. The basisphe noid foramina are dorsoventrally elongate and taper dorsally; they are positioned high within the basisphenoid recess. Contra Russell (1970), the presence of a single median pneumatic fo ramen in the holotype (CMN 8506) is doubtful; the left wall of the recess is crushed medially, obscuring the region of the left foramen completely. In FMNH PR308, the lateral margins of both foramina and a fragment of the median septum between them are visible, the recess in this specimen is otherwise filled with matrix. Further preparation of both specimens is required to clarify structural details.
In Stage 4 D. torosus, the oval scar may bear a strongly rugose surface that is convex in transverse section and passes uninterrupted onto the basal tuber. In FMNH PR308 the oval scar is dished and rugose.
In Stage 4 T. rex specimens (e.g., AMNH 5027), the rostral plate of the bone is rotated caudally to contact the rostral sur face of the basioccipital, eliminating the basisphenoid recess (Fig. 8I). Also, the basisphenoid pneumatic foramina are small and their dorsal margins are at the level of the basioccipital (Fig. 8A, G). The oval scar is concave with a rough surface (Fig. 8 A, G).
In the A. sarcophagus specimens CMN 5600 and TMP 86.64.1, each basisphenoid pneumatic foramen lies within a dis tinct fossa.
Laterosphenoid-In Stage 1 specimens (e.g., ROM 1247) of A. libratus, the caudolateral surface is broadly convex in transverse section and gently excavated ventral to the dorsotem poral fossa. The rostrolateral margin is poorly developed, such that the profundus branch of the trigeminal nerve (N. trigemi nus) (V) exits rostrally (Russell, 1970).
In Stage 4 D. torosus specimens (e.g., CMN 8506), the cau dolateral surface is interrupted by a strong, sharp-edged ledge that separates the dorsotemporal fossa from the rostrolateral surface of the bone. The strut-like rostrolateral margin of the bone displaces the exit for the profundus branch of N. trigem inus (V) caudolaterally (Russell, 1970).
Palatine-In Stage 4 D. torosus (e.g., CMN 8506), the cau dal pneumatic recess is round and there is a broad separation between the recesses. These are also present in the skull of the holotype of A. sarcophagus, CMN 5600. In D. torosus, the palatine is transversely inflated, a condition seen in the A. sar cophagus specimens CMN 5600 and CMN 5601. The palatine is inflated in Stage 4 T. rex specimens (e.g., AMNH 5027).
Surangular-In Stage 4 D. torosus, the surangular may be deep (148; Fig. 4G). The intramandibular process is deep and stout and meets the rostroventral margin of the surangular at a very low angle or is confluent with the rostroventral margin (149; Fig. 4G). The rostral plate is laterally convex (150; Fig.  4G). The dorsomedial flange may be low or dorsally expanded (151; Fig. 4G). The dorsolateral muscle scar may be rugose rostrally. The surangular shelf is depressed over the surangular foramen (152; Fig. 4G). This condition is also in T. rex. The surangular foramen is large and deeply recessed (153; Fig. 4G). The fossa ventrolateral to the glenoid may be dorsoventrally deep or is a rugose pocket (154; Fig. 4G). The caudal margin of the retroarticular process is concave with a caudoventrally projecting heel (155; Fig. 4G).
The surangular of Stage 4 T. rex specimens is deep with a subvertical rostral margin and transversely convex rostral plate (Fig. 8E). The dorsomedial flange is low, and the muscle scar lateral to it is rugose. A deep rugose pocket is present later oventral to the glenoid (Fig. 8E).
Angular-In Stage 1 specimens of A. libratus, the bone ex tends caudal to the surangular foramen (Fig. 5B). The dorso lateral scar passes medial to the caudal plate as a rugose sulcus (156; Fig. 4H). The ventral margin of the rostral process and caudal plate forms a relatively continuous, convex ventral mar gin (157; Fig. 4H).
In Stage 4 specimens of D. torosus, the caudal plate of the bone is dorsoventrally deep (158; Fig. 4I) and may be restricted rostral to the surangular foramen when in articulation (Russell, 1970:fig. 5). It extends caudally in FMNH PR308 (Fig 1; contra Russell, 1970). The dorsolateral scar that passes medial to the caudal flange is strongly pronounced and rugose (159; Fig. 4I). Finally, the rostral process is flexed dorsally relative to the cau dal plate (160; Fig. 4I).
In Stage 4 T. rex specimens, the angular is deep, the rostral process is flexed relative to the caudal plate (Fig. 8E), extends caudal to the surangular foramen ( Fig. 8E; Russell, 1970), and the dorsomedial scar is a pronounced ridge.
Prearticular-In Stage 4 D. torosus, the dorsal margin of the deepened caudal ramus is rostral in position and may be developed into a strong keel (161; Fig. 4L); its dorsal and ven tral margins may be straightened as they taper rostrally (161; Fig. 4L). The distal margin of the paddle-like rostral lamina (162) may be convex (163; Fig. 4M). These features are also present in Stage 4 T. rex (e.g., AMNH 5027).
Splenial-In Stage 1 specimens (e.g., ROM 1247) of A. li bratus, the rostral extent of the articular surface for the dentary is flat and forms a peg-and-socket contact at its rostral extent. In Stage 4 D. torosus (e.g., CMN 8506), the lateral articular surface for the dentary is reinforced by arcuate, interleaving ridges that fit into corresponding slots in the dentary.
Dentary-In Stage 4 D. torosus (e.g., CMN 8506), the an gular process is dorsoventrally deep. In medial view, the sym physis may be contained rostrally between the rostromedially extending lateral surface and caudally by a pronounced bony ridge. The symphyseal surface is lightly rugose. The articular surface for the splenial extends rostrally, indicated by strong rostroventral ridges and slots. Two to four foramina are present (instead of one or two in A. libratus) at the rostral end of the Meckelian canal. The ventral bar beneath the Meckelian foram ina is extremely rugose and transversely convex, obliterating the ventrally passing sulcus. In addition, a low eminence is present at the rostral end of the ventral bar as in T. rex.

Dentition
In Stage 1 and Stage 3 specimens of A. libratus, the first maxillary tooth is incisiform. In Stage 4 D. torosus the first tooth is subincisiform, modified by labiolingual thickening of the tooth. In Stage 1 A. libratus, the maxillary teeth are labio lingually compressed and blade-like. The fourth maxillary tooth has a crown width/length ratio of 0.52 (ROM 1247); in Stage 4 D. torosus, the teeth are thicker-the crown width/length ratio of the fourth tooth is 0.77 (CMN 8506). In D. torosus, the thick maxillary dentition likely increases the depth and width of the alveolar region.
In Stage 1 A. libratus, all of the dentary teeth are labiolin gually narrow, except for the first. The width/length ratio for the crown of the fourth dentary tooth is 0.5 (ROM 1247); the width/length ratio increases caudally, reaching 0.71 at the 14th tooth. In Stage 4 D. torosus, the mesial dentary teeth are labio lingually thick. The crown width/length ratio of the fourth tooth ranges from 0.7 (CMN 8506) to 0.9 (CMN 11594); the distal teeth are also wide, the ratio is 0.73 (CMN 8506).

Ontogenetic Variation in A. libratus
The growth series of A. libratus is divided into three stages; based on the optimal distribution of ontogenetic character states, the growth stages may be characterized as follows. Stage 1 specimens are the least mature, and display nascent ontoge netic states. Stage 2 specimens are typified by the presence of large marginal maxillary neurovascular foramina, a depressed interfenestral strut, an oblique caudal lacrimal suture of the ju gal, a postorbital situated dorsal to the orbit floor, a spheroid occipital condyle, a deep surangular, a large and asymmetrical caudal surangular foramen, a lacrimal cornual process with one apex, a dorsolateral lamina of the lacrimal that is as deep as the antorbital fossa, a ventrally oriented and wide oval scar of the basisphenoid, and a deep scar ventrolateral to the glenoid of the surangular. Also, Stage 2 specimens may exhibit nascent Stage 1 features.
Stage 3 specimens are characterized by an antorbital fenestra in which the height approaches the length, an expansive rostrolateral surface of the maxilla, a convex rostral margin of the rostroventral lamina of the lacrimal, a deep maxillary process of the jugal, and elongate rostral processes of the dorsal ramus of the lacrimal. In Stage 3, all small Stage 1 features are trans formed but large Stage 1 and Stage 2 features may be unmod ified.

Tyrannosaurid Taxonomy
Evidence for ontogenetic changes in the skull and mandible of Albertosaurus libratus (Lambe, 1914) provides parameters by which similar variation in other tyrannosaurid crania may be inferred. As such, inference of a pattern of ontogenetic change in one taxon requires verification in another. The morphological changes in the face, from an early to a late ontogenetic stage, is an alternative hypothesis to interpretations of tyrannosaurid diversity in which sympatric tyrannosaurid taxa are seen as comprising a giant and dwarf, or more lightly built taxon (e.g., Russell, 1970;Molnar, 1980;Currie, 1987;Bakker et al., 1988;Paul, 1988;Carpenter, 1992). This hypothesized pattern re quires further examination.

Nanotyrannus lancensis
CMNH 7541, a damaged skull with lower jaws in occlusion (Fig. 6A-D), was collected from the Hell Creek Formation of Montana in 1942 and described by C. W. Gilmore (1946) in a posthumous publication. CMNH 7541 was heavily restored in plaster, and Gilmore erred in his account of sutural fusion in the skull, a misinterpretation perpetuated by later workers (Rus sell, 1970;Bakker et al., 1988;Paul, 1988). In fact, there is no evidence of sutural fusion in this specimen except for fusion of the intranasal and intraparietal sutures, which is typical of Stage 1 A. libratus. Gilmore noted the similarity of the skull to the smallest specimen of A. libratus then known (AMNH 5664); thus Gilmore (1946) made CMNH 7541 the holotype of the new taxon, Gorgosaurus lancensis.
In his review of tyrannosaurids from western Canada, Russell (1970) referred "G". lancensis to Albertosaurus and accepted Gilmore's interpretation of sutural fusion and its indication of relative maturity. Later, Bakker et al. (1988:17) proposed a new genus of dwarf tyrannosaurid, Nanotyrannus, for "G." lancen sis, offering sutural fusion between the frontal and prefrontal and between the parietal and frontal as the criteria for the adult nature of the specimen.

Relative Maturity of CMNH 7541
The presence of striated cortical bone was demonstrated by Bennett (1993) and Sampson (1993) to distinguish immature, fast-growing individuals from mature specimens for pterosaurs and centrosaurine ceratopsids, respectively. Immature bone grain is lost with increase in size and development of ontoge netic characters, thus providing a crude measure of relative ma turity among reptiles.
On CMNH 7541, immature bone grain is present on the an torbital fossa of the maxilla (Fig. 7A) and lacrimal, lateral sur face of the vomer, dentary, surangular (Fig. 7C), angular, pal atine, jugal (Fig. 7B), ventral process of the maxilla, quadra tojugal process of the squamosal, squamosal ramus of the post orbital, rostral surface of the supraoccipital crest of the parietal, medial surface of the prearticular and splenial, caudal margin of the quadratojugal, caudal surface of the quadrate, and dorsal surface of the frontal and nasals (Fig. 7D).
Premaxilla-In rostral view, the premaxillae are narrow and their lateral margins are concave at the base of the maxillary processes, and the alveolar region is shallow (Fig. 8J).
Maxilla-The maxillae of CMNH 7541 are laterally flat tened, the alveolar process is shallow, the first tooth is incisi form, and the remaining teeth are labiolingually narrow (Fig.  6A, D). Also, the ventrolateral rim of the ventral jugal process is not breached by the caudalmost neurovascular sulcus of the ventral row of foramina (Fig. 6A). The lateral surface does not extend caudally over the rostral margin of the antorbital fossa (Fig. 6A), the promaxillary fenestra is dorsoventrally elongate and not recessed (Fig. 6A), the antorbital fenestra is longer than high, and the small maxillary fenestra is midway between the rostral margins of the antorbital fossa and restored fenestra (Fig.  6A).
Nasal-In CMNH 7541, the nasals are smooth with low transverse and fine rostrocaudal ridges (Fig. 6A, B). In the smallest Stage 1 specimen of A. libratus examined (TMP 86.144.1), the rugose texture typical of mature specimens is present. The condition in CMNH 7541 might represent individ ual, ontogenetic, or taxonomic variation.
Lacrimal-In CMNH 7541 the rostral margin of the ros troventral lamina is concave to straight and the contact of the lamina with the jugal exceeds that of the ventral ramus as in Stage 1 A. libratus (Fig. 6A). Also, there is no evidence of fusion between the ventral margins of the medial and lateral processes of the rostral ramus.
Jugal-As in Stage 1 A. libratus, the maxillary ramus of CMNH 7541 is dorsoventrally shallow and tapered (Fig. 6A). The jugal pneumatic recess is a rostrally-restricted slit (Fig.  6A), the postorbital articular surface approaches the orbit floor (Fig. 6A), the region ventral to the postorbital ramus is convex (Fig. 6A), the caudal margin of the postorbital ramus is convex at midheight, and the caudal rim of the lacrimal articular surface is subvertical (Fig. 6A).
Postorbital-As in Stage 1 A. libratus, the laterodorsal mar gin is not everted medially into the dorsotemporal fenestra (Fig.  6A, B). There is no postorbital cornual process or suborbital prong (Fig. 6A); the former is represented by a textured surface. Unlike A. libratus, the rostral and caudal margins of the jugal process taper rostroventrally (Fig. 6A).
Prefrontal-As in Stage 1 A. libratus specimens, the pre frontal is situated at the rostrolateral margin of the frontal, bounded caudally by a triangular, tab-like process from the frontal to separate it from the lacrimal caudolaterally (Fig. 6B).
Frontal-As in Stage 1 A. libratus, the lacrimal notch in CMNH 7541 is elongate and narrow in dorsal view, the pre frontal notch is rostrocaudally stout in dorsal view, and the paired frontals are longer than wide (Fig. 6B). Further evidence of the immature nature of the specimen is provided by the rel atively elongate orbital margin (11 mm) and the shallow dor sotemporal fossa, which has a barely discernible rostral margin (Fig. 6B). Unlike Stage 1 A. libratus, the frontals appear to rise to meet each other along the midline rostral to the dorsotem poral fossa (Fig. 6B); however, cracks suggest this may be an artifact of dorsoventral crushing.
Parietal-In CMNH 7541 the delicate nuchal crest is low in caudal view (Fig. 6C). Its dorsal margin is rostrally everted and the laterodorsal margin is convex in frontal section (Fig. 6A, B). Unlike Stage 1 A. libratus, the sagittal crest in lateral view is tall and blade-like.
Supraoccipital-As in Stage 1 A. libratus, the dorsal process of the supraoccipital of CMNH 7541 is narrow and has a hor izontal dorsal border with flange-like rostrolateral edges (Fig.  6C).
Basioccipital-As in Stage 1 A. libratus, the ventrolateral margins of the occipital condyle taper toward each other ven trally and the caudoventral condylar surface is flattened (Fig.  6C). The ventral surface of the basituberal web is flat and arched in caudal view (Fig. 6C). The median portion of the basioccipital is concave between the laminae and the bone is strongly excavated caudolaterally by the subcondylar recess (Fig. 6C).
Parabasisphenoid-As in Stage 1 A. libratus, the pneumatic foramina in the basisphenoid are small and situated ventrally and the oval scar is smooth and lateroventrally oriented (Fig.  6C, D).
Dentary-As in Stage 1 A. libratus the dentary of CMNH 7541 is relatively shallow in lateral view and narrow in ventral view (Fig. 6A, D). Surangular-In CMNH 7541 the surangular is shallow and the surangular shelf is horizontal (Fig. 6C).
Prearticular-As in Stage 1 A. libratus the dorsal margin of the caudal ramus of the prearticular is restricted caudally (Fig. 6D), the caudal ramus is shallow with parallel dorsal and ventral margins (Fig. 6D), the rostral lamina is slightly expand ed and its parallel margins converge to a point distally (Fig.  6D), and the caudodorsal surface of the rostral lamina is smooth (Fig. 6D).

Immaturity of CMNH 7541
The morphological structure of CMNH 7541 agrees with that of Stage 1 specimens of A. libratus and displays no mature features. The weight of evidence indicates that CMNH 7541 is juvenile. The presence of immature bone grain (Fig. 7) pre cludes the specimen from representing an adult pygmy tyran nosaurid.
Status of Nanotyrannus- Rozhdestvensky (1965) first sug gested that CMNH 7541 might be a juvenile Tyrannosaurus rex, based on his observations of T. bataar, in which juveniles vary in "slightly different proportions than in the large speci mens, as we should expect if they are of different (ontogenetic) ages" (Rozhdestvensky, 1965:106). Carpenter (1992) also suggested that CMNH 7541 might be an immature specimen of T. rex on the basis of its long and low snout, separate nasals and frontal, circular orbit, and ros trocaudally oval margin of the antorbital fenestra. Carpenter was uncertain of the significance of the last feature, owing to its acute rostral margin relative to that in immature specimens of T. bataar. In fact, on both sides, this region is absent and restored in plaster in CMNH 7541. Carpenter's suggestion con cerning ontogenetically variable features, except for the nasalfrontal contact, is consistent with the observations reported here. He noted two features that united the specimen with Ty rannosaurus: considerable lateral constriction of the snout, and a dorsally broad temporal region (Fig. 8C, D). However, Car penter (1992:528) tentatively accepted Russell's identification of CMNH 7541 as ?A. lancensis.

Ontogeny in Tyrannosaurus rex
Ontogeny in the craniofacial skeleton of Tyrannosaurus rex is characterized by a global shift from a gracile early ontogeny to a robust late ontogeny morphotype. In T. rex the rostral max illary and dentary teeth become conical, expanding and deep ening the alveolar processes of the maxilla and dentary (Fig.  8A, C, E, F). The tooth row becomes rostrodorsally reoriented, and the teeth become procumbent (Bakker et al., 1988). Also, the maxilla loses three to four teeth from the rostral end of the tooth row, because the rostral teeth undergo the greatest change. Although tooth count has been used to distinguish tyranno saurid taxa (e.g., Bakker et al., 1988), caution is advised be cause counts appear to be sensitive to ontogenetic and individ ual variation.
A similar pattern of loss of tooth positions is present in the maxilla of A. libratus, dropping from 16 to 13 alveoli (Table  2). Although one large Stage 1 specimen (USNM 12814) has a low alveolus count, this specimen is the most mature of the growth stage (Carr, in prep.). While loss of alveoli may be individually variable, it is evident that the phenomenon occurs in mature specimens (Table 2). Among other archosaurs, on togenetic tooth loss has been reported by Mook (1921), Wermuth (1953), andIordansky (1973) in Crocodylus cataphractus, C. porosus, C. siamensis, and Tomistoma schlegelii. This in dicates that ontogenetic tooth loss among Tyrannosauridae is not withXout precedent among Archosauria.
In agreement with Molnar's detailed functional analysis (1973), the adult skull of T. rex is constructed for forceful bit ing. In T. rex, the dorsotemporal fossa becomes deeply exca vated during ontogeny, reflecting hypertrophied adductor mus culature (Fig. 8C, D). The entire skull is modified to accom modate the change, including the rostrally-oriented orbits (Fig. 8C,D,I,J). In concert with the increase in bite force, the muzzle and jaws become deep and contacts between bones are strengthened by peg-and-socket sutures, such as the nasomax illary contact (Fig. 8E). The facial skeleton is buttressed to de liver and absorb the resultant forces of increased bite force, evidenced by the strut-like rostral margin of the antorbital fossa, which passes to the columnar dorsum of the snout. Also, the premaxillary dental arcade of early growth stages is abandoned for a sparser arrangement of teeth (Fig. 8I, J), indicating the less specialized grasping function characteristic of the remain der of the rostral maxillary tooth row.
In addition, the craniofacial air sac system (sensu Witmer, 1987Witmer, , 1990Witmer, , 1997) had a primary role in modifying bone struc ture. The antorbital air sac rested within the antorbital fenestra and antorbital fossa, sending diverticula into the ectopterygoid, palatine, lacrimal, jugal, and maxilla. As maturity was attained, the diverticula invaded the bones more fully, expanding the sinuses and bones. Combined with changes induced by an en larged dentition, late ontogeny specimens became 'swollen faced' relative to their smaller, sleeker progeny. It is probable that the swollen bones had greater cross-sectional strength than the strap-like bones of smaller animals, a morphological shift that would be important for taking live prey.
Finally, pneumatic features indicate further ontogenetic  (Bakker et al., 1988) 678 (Russell, 1970 It is evident that the distinct structural patterns of early and late ontogeny specimens of T. rex constrained their respective foraging behaviors, which in late ontogeny individuals appears specialized to grasp and hold live prey or to dismember a large carcass. Alternatively, the changes represent a reaction norm to the size of the skull, increased bite forces, and increased prey size. Nonetheless, it is not improbable that small and large an imals differed in foraging strategy, consumption technique, and/ or prey type. Such size dependent differences are found in ex tant crocodilians (Grenard, 1991) and monitor lizards (Auffenberg, 1988(Auffenberg, , 1994.

Maleevosaurus novojilovi and Tyrannosaurus bataar
PIN 552-2 was originally described by Maleev (1955a) as G. novojilovi. Rozhdestvensky (1965) recognized this specimen as a juvenile of T. bataar. In 1992, Carpenter proposed the new genus, Maleevosaurus for "G." novojilovi, arguing that the small maxillary fenestra, promaxillary fenestra not visible in lateral view, large antorbital fenestra, low lacrimal horn, low postorbital cornual process, shallow maxilla, and slender jugal fall outside of the range of variation seen in juveniles of A. libratus and T. bataar. Maleev's published figure (1974:fig. 55) of the skull indicates that PIN 552-2 conforms to the features characteristic of early ontogeny A. libratus'. delicate nasal with a slotted maxilla su ture, long antorbital fenestra, round maxillary fenestra posi tioned midway between the interfenestral strut and rostral mar gin of the antorbital fossa, promaxillary fenestra not recessed, rostral margin of the antorbital fossa not overlapped by the lateral surface, shallow alveolar process of the maxilla, shallow and delicate jugal (Carpenter, 1992), low postorbital cornual process (Carpenter, 1992), and shallow dentary (Carpenter, 1992).
As in late ontogeny specimens of T. bataar (Maleev, 1974: figs. 1, 48), there is no lacrimal cornual process, the maxillary fenestra is relatively large, and the postorbital cornual process is patch-like. Because of the incompleteness of the specimen, the number of diagnostic features to be gleaned from an illus tration is limited. However, in the absence of apomorphies and on geographic and stratigraphic grounds, it is most parsimoni ous to consider PIN 522-2 as a young specimen of T. bataar, as first suggested by Rozhdestvensky (1965).

Overall Conclusions
Tyrannosaurid craniofacial ontogeny follows a conservative pattern that can be observed across taxa. The recognition of ontogenetic variation reduces the number of sympatric taxa and renders the concept of dwarf tyrannosaurids untenable. Imma ture tyrannosaurids have been misidentified as pygmy adults (e.g., Maleev, 1974;Gilmore, 1946;Bakker et al., 1988;Car penter, 1992). Small skulls and mandibles are delicate in con trast to the robust large specimens.
An additional source of confusion is the resemblance of early ontogeny tyrannosaurid skulls to those of adult small theropods. A tyrannosaurid skull 37% of maximum length has an orbit 17% of its skull length. Adult small theropods have similar, if not identical, proportions (e.g., Coelophysis bauri, CM C-4-81; skull length 250 mm; orbit/skull length ratio 16%) (Colbert, 1989). In contrast, in small theropods, small crania 35% of maximum adult length have relatively enlarged orbits (e.g., C. bauri, MCZ 4326; skull length approx. 88 mm; orbit skull length ratio 34%) (Colbert, 1989). The comparable proportions of early ontogeny tyrannosaurids with adult small theropods may have misled previous workers. Although the immature tyr annosaurids examined herein are not hatchlings, it is perhaps noteworthy that they lack the enlarged orbits and short snouts of juvenile small theropods. It is possible that hypotheses con cerning particular modes of heterochrony might be generated to explain the phenomenon. In this case, acceleration and hypermorphosis might be invoked (sensu Reilly et al., 1997). (1) nasal processes of the premaxillae tightly appressed throughout their entire length; (2) restricted exposure of the jugal within the antorbital fenestra; (3) antorbital fossa reaches the nasal suture caudodorsally; (4) transversely broad jugal pneumatic recess; (5) elongate frontal sagittal crest; (6) strongly divergent and short basal tubers; (7) rostroventrally-oriented caudal occipital plate; (8) shallow subcondylar recess; (9) rostroventrally deep basisphenoid plate and rostrocaudally-restricted basisphenoid recess; (10) inflated ectopterygoid; (11) strongly convex rostral plate of the surangular; (12) transversely narrow snout and broad temporal region relative to other tyrannosaurids; and (13)