BACKGROUND: The long-term predictive ability of the ICD-10 subtypes of hypomania, mania without psychosis and mania with psychosis has not been investigated. METHODS: All patients who got a diagnosis of a single hypomanic episode, single manic episode without psychosis or single manic episode with psychosis in a period from 1994 to 1999 at the first discharge ever in Denmark were identified. The time to discharge from first admission and the risk of relapse leading to readmission were compared for the three groups of patients. RESULTS: Totally, 41 patients with a hypomanic episode, 149 patients with a manic episode without psychotic symptoms and 202 patients with a manic episode with psychotic symptoms at first discharge ever were identified. Patients with mania and psychotic symptoms were admitted longer than patients with mania without psychosis, and patients with mania without psychosis were admitted longer than patients with hypomania. No differences were found in the risk of relapse leading to readmission between the three groups of patients. LIMITATIONS: The results apply to hospitalised patients only. CONCLUSIONS: The ICD-10 subclassification of manic episodes does only partially predict long-term outcome.