The purpose of this study was to clarify the issue of whether DSM-III-R (American Psychological Association [APA], 1987) over- or underdiagnoses autism by comparing this diagnostic system to a well-established objective measure of diagnosis, the Childhood Autism Rating Scale (CARS). A secondary goal was to determine which of the 16 criteria are the best discriminators of autism. DSM-III-R, CARS, and clinical diagnoses of 138 consecutive admissions to a statewide program for the diagnosis and treatment of autistic and related communication-handicapped individuals (Division TEACCH in North Carolina) were compared. Results indicated a generally high degree of agreement on the diagnosis of autism using the three systems. Within this treatment-oriented program, the CARS and clinical ratings diagnosed a greater number of cases as autistic than did the DSM-III-R criteria, suggesting that DSM-III-R slightly underdiagnosed autism. The criteria that most strongly related to the diagnosis of autism regardless of the system were lack of awareness of others, abnormal social play, an impaired ability to make friends, abnormal nonverbal communication, stereotypic body movements, and restricted range of interests.