In the present study, factor mixture models (FMMs) were used to examine the latent structure underlying the Autism-Spectrum Quotient (AQ) among a sample of 633 undergraduate students. FMM represents a combination of latent-class, person-centered approaches and common-factor, variable-centered approaches to modeling population heterogeneity. Findings suggest the presence of either two or six latent classes with distinct profiles across the set of 50 AQ items. In addition, within each class, individuals can be further differentiated according to their scores on five latent factors. These results suggest the presence of phenotypical heterogeneity at the sub-clinical level in addition to that which is known to exist at the clinical level.