The purpose of this study was to investigate diagnostic agreement between clinicians and a research group in a sample of first-admission psychosis and severe affective disorder patients. Clinical DSM-IV discharge diagnoses and best-estimate DSM-IV research diagnoses were compared in 116 first-episode patients in the city of Turku, Finland. The best-estimate research diagnoses were made at consensus meetings by integrating longitudinal data; patients' medical records; and findings of a clinical interview, the structured SCAN-interview, and symptom severity ratings. Overall diagnostic agreement was moderate, with a kappa value 0.51 (95% confidence interval (CI), 0.39- 0.63). Of the diagnostic groups, schizophrenic disorders had the lowest kappa value of 0.44 (95% CI, 0.26-0.63). Clinicians had a tendency to miss depressive symptoms in psychotic patients; to overdiagnose psychotic symptoms in depressive patients; and to fail to discover earlier hypomanic or depressive episodes in depressive patients. In conclusion, hospital diagnoses were not reliable in first-episode patients. Inappropriate diagnoses may compromise both treatment and epidemiologic findings based on discharge diagnoses.