A multivariate approach using pattern recognition method was applied on a multivariable data set from patients with affective disorders comprising biological and clinical variables. The depressed patients were rated according to 23 items of the comprehensive psychopathological rating scale (CPRS). Variables of importance were selected and clusters of patients were found by combining monoamine oxidase, melatonin and post-dexamethasone cortisol with symptoms of psychomotor retardation and agitation. Patients were distributed with high scores of agitation in the extreme of one direction and with high scores of retardation in the opposite direction. By using the combined clinical and biological variables, a diagnostic subcategory with latent bipolar disorder was identified. Two clusters of unipolar patients, one with low melatonin and low psychomotor retardation scores, and one with high melatonin and high psychomotor retardation scores, were found. Identification of a patient group with latent bipolar disorder may have potential therapeutic value since bipolar patients should be taken care of by a specialist in psychiatry, avoid tricyclic antidepressant therapy and may be candidates for lithium treatment.