A transfer of stimulus control procedure was used to teach three profoundly retarded adolescents a series of specific responses to specific verbal instructions. After imitative control of a behavior was established, a verbal instruction was presented immediately before the behavior was modelled. Each correct response was followed on the next trial by inserting a delay between the verbal instruction and the modelling of the behavior. The delays increased from trial to trial. Transfer of stimulus control was indicated when a subject responded correctly on five consecutive trials before the behavior was modelled. All three subjects responded correctly to each verbal instruction after that item was trained in a multiple-baseline order. Generalization did not occur to items that had not been trained. Probe data revealed that some variations of the verbal instructions controlled responses after training was completed.