Comparative investigations of mandibular function among primates have relied upon elementary structural models to estimate states of masticatory stress and strain. In these studies, mandibular corpus morphology is idealized as a homogeneous, isotropic symmetrical body of invariant geometry, and this morphological abstraction is used to infer relative levels of stress and strain in the jaw. In reality, none of the limiting conditions assumed by these models is satisfied; consequently, it is prudent to ask whether this "textbook" engineering approach is valid for the inference of biomechanical behavior. In this study, the predictions of various geometric representations of the mandibular corpus are evaluated against strains recorded in a sample of human jaws loaded in torsion. Symmetrical geometrical models (including various "robusticity" shape indices), although convenient, are probably not consistently reliable for predicting the distribution of strains in the corpus. The experimental data suggest that variations in cortical thickness within sections play a significant role in determining the profile of relative strains. For comparative applications, characterization of the corpus as an asymmetrical hollow ellipse (i.e., with differing thickness of medial and lateral cortical plates) may provide a reasonable portrayal of relative strains.