To determine the best discriminative items for identifying young children with autism spectrum disorders (ASD), we conducted a secondary analysis using longitudinal cohort data that included the Japanese version of the 23-item modified checklist for autism in toddlers (M-CHAT-JV). M-CHAT-JV data at 18 months of age and diagnostic information evaluated at age 3 or later from 1851 Japanese children was used to isolate six highly discriminative items. Using data from two different community samples (n = 1851, n = 665) these items were shown to have comparable psychometric values with those of the full version. Our results suggest that these items might work as a short form screener for early identification of ASD in primary care settings where there are time constraints on screening.