Six severely mentally retarded students (ages 9-16) were paired with six nonhandicapped fourth- and fifth-graders (Group 1) who received training to act as tutors during outdoor recess. Six of their nonhandicapped friends (Group 2) received no training. Group 3 consisted of all other fourth- and fifth-graders who were also present during the recess period. Social interaction data were collected for the nonhandicapped students on five dependent variables: active participation, social initiation, responding, teaching, and reinforcing. Three independent variables (social interaction training for Group 1, feedback, and followup) were systematically introduced during intervention. Results indicated that, as training progressed, the frequency of the trained behaviors for Group 1 substantially increased. Effects of training generalized to Group 2, and Group 3 showed an overall increase in social interactions with the handicapped students. It is suggested that training a few subjects could produce a generalization effect, and that unless the nonhandicapped children are trained, there will not be any high level of interaction between handicapped and nonhandicapped students. Appended are a student handbook for peers of handicapped students, an accompanying training guide for teachers or trainers, and a section on data collection procedures. (JW)