A method is presented for empirical subclassification of autistic and autisticlike children, based on observations of current behavior. The advantage of the method is that it identifies profiles of co-occurring behaviors and accordingly assigns children to subtypes. The subtypes are more clinically homogeneous than the overall sample of autistic children. Preliminary findings are presented, including an effort to validate the subclasses by suggesting possible relationships between subtype membership and perinatal markers, developmental milestones, and independent measures of concurrent behavior.