This study extended previous investigations of autistic subjects' perspective-taking abilities through a within-subjects contrast between two levels each of both visual and cognitive problems with stringent controls against guessing. When compared with normal and mentally retarded subjects', the autistic group's performance supported Baron-Cohen's (1988) hypothesis of a selective deficit for cognitive perspective taking among autistic subjects. Both levels of visual perspective taking demonstrated virtually unimpaired performance for autistic subjects with no significant difference between them and control groups. On the cognitive perspective-taking tasks, however, the performance of the three groups was significantly different, with the vast majority of autistic subjects unable to do even the most basic level of this task. Possible explanations and educational implications were discussed.