Learning set, matching learning set and conditional matching learning set tasks were presented to matched groups of autistic, retarded and normal children. The majority of autistic children were able to solve these problems and to acquire set and their performance was similar to that of MA matched controls. It is suggested that (a) performance was directly related to developmental level rather than to autism per se, (b) that learning was enhanced by the use of concrete stimuli since conditional cues were successfully used here, in contrast to failure in an earlier experiment using abstract cues, and (c) overselectivity was not characteristic of the performance of these autistic children.