Generalization across three stimulus parameters was examined for 5 individuals whose self-injurious behavior was maintained by escape from task demands. Prior to treatment, three stimulus parameters (therapist, setting, and demands) were systematically varied across baseline sessions. These variables were held constant during treatment, which consisted of escape extinction. When treatment was completed, three novel stimulus parameters were probed. If the rate of self-injury was high during this probe, treatment was reimplemented with one new stimulus parameter (the other two were the same as in the original treatment condition). Following this second treatment, another probe with three novel stimuli was conducted. If the rate of self-injury was again high, treatment was implemented again while a second stimulus parameter was changed. This sequence continued until generalization was observed across the three parameters. Results showed idiosyncratic differences in generalization. The behavior of 2 subjects showed complete generalization during the first novel probe. A 3rd subject's behavior showed generalization following treatment across two stimulus parameters (setting and therapist). The behavior of the 2 remaining subjects showed a complete lack of generalization across the three parameters; both subjects required training for novelty by randomly varying the stimulus parameters for a substantial number of sessions.