Seventy-seven 10-18-year-old psychiatric in-patients and out-patients took part in a test-retest study of the Child and Adolescent Psychiatric Assessment (CAPA). They were interviewed on two occasions several days apart. Overall reliability of diagnosis ranged from kappa = 0.55 (conduct disorder) to 1.0 (substance abuse or dependence). In general, reliability for scale scores of psychopathology was somewhat lower in out-patients than in-patients, though the opposite was the case for anxiety disorders and psychosocial incapacity and the reliability of the diagnosis of conduct disorder--the only individual diagnosis sufficiently common to permit this comparison. Unreliability of reports of behavioural problems was found to be related to admitting to being a liar in the first interview. The implications of these results for the use of the CAPA are discussed.