Results from ROC curves of items from two scales, the Autism Diagnostic Interview (ADI) and Autism Diagnostic Interview-Revised (ADI-R), operationalizing DSM-IV criteria for autism are presented for 319 autistic and 113 other subjects from 8 international autism centers. Analyses indicate that multiple items were necessary to attain adequate sensitivity and specificity if samples with varying levels of language were considered separately. Although considering only current behavior was generally sufficient when a combination cutoff and additive model was employed, predictive power was highest when history was taken into account. A single set of criteria, as operationalized by individually structured questions in the ADI/ADI-R, was effective in differentiating autism from mental handicap and language impairment in subjects with a range of chronological ages and developmental levels.