Twenty-two children with autism were given four tests of false belief understanding: the Sally-Anne task, two variants of the deceptive box task, and the three boxes task. The overall consistency of the children's performance was high, 77 percent of the participants either passing or failing all of the tasks. The convergent validity (across-task consistency) of the deceptive box and the three boxes paradigms was high, and the convergent validity of the three boxes and Sally-Anne tasks was also acceptable. However, a weaker level of convergent validity was found for the deceptive box and Sally-Anne tasks, suggesting that these paradigms test slightly different aspects of cognition. The reliability (within-child consistency) of the children's performances across two versions of the deceptive box task was high. These findings are discussed in terms of their practical implications for practitioners and researchers.