The validity of the self-administered CIDI-Auto for detecting ICD-10 diagnoses was assessed in a study of 126 patients admitted to an acute psychiatry unit. A comparison was made between the level of agreement of the CIDI-Auto with a psychiatrist and that between two psychiatrists. The CIDI-Auto generated an average of 2.3 diagnoses per subject, and the psychiatrists 1.3. Agreement measured by overall agreement and by Kappas between the CIDI-Auto and the psychiatrist's principal diagnosis was poor, whereas agreement between psychiatrists was good. At the level of general diagnostic class (e.g. substance use disorder, schizophrenic disorder, mood disorder), agreement between CIDI-Auto and psychiatrist on principal diagnosis was poor, Kappa = 0.23, while agreement between psychiatrists was good, Kappa = 0.69. The findings indicate that the self-administered CIDI-Auto has poor validity measured against clinical diagnosis for hospitalised patients of acute psychiatric services. Poor validity of computer-based diagnosis limits the diagnostic utility of these methods in clinical situations. It also creates uncertainty of diagnostic findings in survey use.