Seven mentally retarded adolescents received either verbal or nonverbal training on a shape-sorting task within a multiple baseline design. Three clients received the nonverbal training sequence, which used contingent attention and praise for correct responses and a correction procedure for incorrect responses. The remaining four clients received the verbal training sequence, which was the same as the nonverbal sequence except that the clients were taught to verbalize (label) the relevant dimension of the objects they were sorting. Both training sequences resulted in the successful acquisition of the target performance, but only the verbal training sequence facilitated generalization of performance to a novel shape and to a new dimension. Following generalization testing, clients initially exposed to the nonverbal sequence were provided verbal training, which produced generalization in these previously unsuccessful clients.