Summary. Previous research on the child's ability to predict the viewpoint of another has typically used either (a) a recognition procedure, in which the subject views the model from different perspectives prior to selecting the viewpoint required, or (b) an inference procedure, in which the subject makes his choice after viewing the model from one position only. An experiment is reported which directly compares the effects of these two procedures, and demonstrates less egocentrism in a recognition than in an inference condition. It is suggested that this distinction can help explain some apparent anomalies in the results of earlier research.