The effects of a graduated prompting treatment procedure were analyzed in three phases of an experiment on the training and generalization of reaching-grasp responding in 2- to 4-year-old blind, severely or profoundly retarded children. In Phase 1, we used a multiple-baseline across-subjects design to investigate the effectiveness of the treatment on midline reach-grasp responding. In Phase 2, we used a reversal design to investigate the effects of repeated implementations and withdrawals of the treatment. In both phases, generalization to right and left positions was measured. In Phase 3, in a multiple-baseline across-responses design, the treatment was implemented in right and left positions. Also in Phase 3, shift of stimulus control from toy-sound to verbal instructions was measured. The results showed that the graduated prompting procedure was effective in training reach-grasp responding in all three children; for one child, the effects were durable over repeated applications of the treatment procedure, but were not maintained during withdrawals; for another child, the treatment procedure was effective in teaching reach-grasp responding in all three positions; and for the same child, training of reach-grasp responding generalized to toys presented without sound, given only the verbal instruction.