After more than three decades of studies, video is still a vaguely understood medium for training. In this study, self-modeling (self-observation of videotapes that show only adaptive behavior) was compared with videotaping only, using a within-subject type of design. Subjects were 18 boys and girls aged 5 to 13 years with various disabilities including cerebral palsy and spina bifida. Each child was assigned two target behaviors for potential intervention. One behavior was treated with video self-modeling and the other was videotaped without further intervention, resulting in a significant treatment effect. Self-model recordings were produced by planning and selectively editing two minutes adaptive-oly behavior, which subjects reviewed on six occasions over two weeks for a total of 12 min intervention. Progress was confirmed one year later. The study supports the efficacy of self-modeling for selected behaviors of these children with physical disabilities, and suggests further investigation of structured video replay as an active agent of change.