We evaluated the effects of a peer-training strategy, consisting of direct prompting and modeling, on the occurrence and duration of interactions between autistic students and nonautistic peer-trainers. Data were obtained in both training and generalization settings. The results of a multiple-baseline design across students demonstrated that:the direct prompting procedure produced immediate and substantial increases in the occurrences and durations of positive social interactions between the peer-trainers and autistic students; these increases were maintained across time at levels above baseline during subsequent free-play probes; these findings were judged by teachers to be socially valid; untrained peers increased their interactions with the autistic students in three of the four groups; generalization of behavior change across settings occurred only after specific programming; and interactions between untrained peers and peer-trainers decreased following training. Variables that may account for the results and the implications of these findings for peer-mediated interventions are discussed.