The present study evaluated the hypothesis that autism spectrum disorders (ASDs) are best represented as a discrete category distinct from typical behavior within autism-affected families. The latent structure, categorical versus dimensional, of ASDs informs future diagnostic revisions, clinical assessment, and the design of future research. Data were obtained from Interactive Autism Network, a registry that preferentially recruits families with at least one ASD-affected child. Caregivers reported autism symptoms for affected and unaffected children using the Social Responsiveness Scale and Social Communication Questionnaire. Taxometric and latent variable models examined whether dimensional or categorical models best fit the data. Taxometric and latent variable model comparisons consistently indicated two-group mixtures for all indicator sets, even in participants designated as unaffected by caregivers. The identified category was associated with external indicators of disability, supporting its validity. Results indicated that ASD is best characterized as a category, distinct from typical behavior within ASD-affected families. (Contains 3 tables and 2 figures.)