A specially designed computer version of the Sally-Anne false task belief task was used to teach understanding of false belief to three groups: children with autism, children with Down's Syndrome and young normal children. In an initial assessment children were selected for teaching only if they failed four false belief tasks: the dolls version of the Sally-Anne task (close transfer task) and three other false belief tasks involving different scenarios (distant transfer tasks). Following teaching, all three groups were able to pass the Sally-Anne task, but the children with autism alone were unable to pass the distant transfer tasks. The possibility that the children with autism had developed an alternative strategy in order to pass the instruction task is discussed.