Solving belief problems develops as a skill in normal children during the preschool years. To understand this process of development, it is necessary to provide an analysis of the tasks used to test preschool 'theory of mind' skills. This analysis should allow us to relate the structure of a given task to the underlying cognitive mechanisms that the task engages. In two experiments, we find that 3-year-old children show a pattern of success and failure on belief tasks that is not consistent with 'conceptual deficit' accounts. Young children possess the concept, BELIEF, but have certain characteristic difficulties with correctly calculating the contents of beliefs. In childhood autism, by contrast, the mechanisms that in normal development bestow conceptual competence in this domain are impaired. In the first experiment, parallel task structures are used to show that 3-year-olds are no better at predicting behavior from a partially true belief than they are at predicting behavior from an entirely false belief. We develop specific proposals about task structural factors that either facilitate or hinder success in belief-content calculation. These proposals are supported in a second experiment. We compare two false-belief tasks, one of which has helpful structural factors, the other of which has hampering factors, with a third task which exemplifies a hampering task structure but without any theory of mind content. We compare 3- and 4-year-olds' patterns of performance with that of autistic children. Each of the three groups shows a distinct performance profile across the three tasks, as predicted for each case by our model. Innate attentional mechanisms provide the conceptual foundations for 'theory of mind' but must be supplemented by a robust executive process that allows false beliefs to achieve 'conceptual pop-out.' Our approach has general implications for the study of conceptual development.