Four chronic mental patients, residents of a token economy treatment unit, were randomly assigned to an experimental or control condition and attended 10 thirty-minute treatment sessions. For the two experimental subjects, contingent reinforcement was received for interacting with each other according to instructions in four distinct phases of interpersonal behavior: talking to another person, attending and talking to another person, asking and answering questions in a dyad, and working cooperatively in the dyad to solve problems. The two control subjects were instructed to perform the same behaviors but received non-contingent reinforcement. The results indicated a strong contingent reinforcement effect on the performance of the treatment sessions' target behaviors. Several baseline, treatment, and post-treatment response measures indicated that the treatment effects had generalized to other areas of social behavior away from the treatment setting.