Experimental figures and labels were selected that theoretically approximate the difficult discriminations encountered by young children in school (e.g., b vs d). Following pretraining on three figures, the children were randomly assigned to one of four experimental groups to investigate different methods of integrating a new, similar symbol into a set of familiar symbols. The new symbol was visually similar and its label auditorily similar to one of the three familiar symbols. The results supported Carnine's (1976) hypothesis that similar correspondences should be separated from each other whenever possible. That is, training conditions that provided for initial separation of similar elements (either in shape or name), required fewer training trials than conditions that did not provide such separation. Teaching implications are discussed.