We prospectively examined mean changes in Autism Diagnostic Interview-Revised (ADI-R) Total and Domains scores and stability of the ADI-R diagnostic classification in 28 children with autism initially assessed at age 2-4 years and reassessed 2 years later. Mean Total, Social Interaction, and Communication scores decreased significantly from Time 1 to Time 2 Restricted/repetitive Domain mean scores did not change over time. The ADI-R diagnostic classification was stable in 67% of children using the current published criteria. The stability increased to 78% when a modified criterion was used in the Restricted/repetitive Domain and to 88% when the broader ASD criteria were used. Among pre-schoolers with autism, parent-reported symptoms decreased significantly at two-year follow-up in Social and Communication Domains but not in the Restricted/repetitive Domain. However, ADI-R diagnostic classification remained relatively stable over time. Revising ADI-R diagnostic criteria in the Restricted/repetitive Domain or including the broader ASD criteria may improve its sensitivity and diagnostic stability in younger children. (Contains 2 tables and 1 figure.)