Previous studies of cranial shape have established a consistent interspecific allometric pattern relating the relative lengths of the face and braincase regions of the skull within multiple families of mammals. In this interspecific allometry, the facial region of the skull is proportionally longer than the braincase in larger species. The regularity and broad taxonomic occurrence of this allometric pattern suggests that it may have an origin near the base of crown Mammalia, or even deeper in the synapsid or amniote forerunners of mammals. To investigate the possible origins of this allomeric pattern, we used geometric morphometric techniques to analyze cranial shape in 194 species of non-mammalian synapsids, which constitute a set of successive outgroups to Mammalia. We recovered a much greater diversity of allometric patterns within non-mammalian synapsids than has been observed in mammals, including several instances similar to the mammalian pattern. However, we found no evidence of the mammalian pattern within Therocephalia and non-mammalian Cynodontia, the synapsids most closely related to mammals. This suggests that the mammalian allometric pattern arose somewhere within Mammaliaformes, rather than within non-mammalian synapsids. Further investigation using an ontogenetic series of the anomodont Diictodon feliceps shows that the pattern of interspecific allometry within anomodonts parallels the ontogenetic trajectory of Diictodon. This indicates that in at least some synapsids allometric patterns associated with ontogeny may provide a “path of least resistance” for interspecific variation, a mechanism that we suggest produces the interspecific allometric pattern observed in mammals.