The purpose of this investigation was to compare two instructional models designed to teach autistic children a bowling task. One strategy (referred to as the extra-stimulus prompt model) used extensive physical, visual, and verbal prompts while the second (referred to as the within-stimulus prompt model) minimized such prompts. With the theory of overselectivity, it was predicted that the within-stimulus prompt model would be the more effective. Both instructional models included a 14-level task analysis of bowling. Subjects were 6 autistic boys between the ages of 7 and 10 years. Group and time series designs were utilized; 3 subjects in each condition performed 332 trials of the task. The dependent variable was improvement on the bowling task as demonstrated by the task analytic level achieved by each subject. The student-teacher interaction was videotaped and assessed for number and types of prompts, reinforcement, and punishment. Nonparametric and visual analyses revealed that the extra-stimulus prompt group performed significantly better in bowling than did the within-stimulus prompt group. No differences occurred in reinforcement or punishment received.