We describe a double-blind study involving 58 in-patients with major depression (DSM-III). After one week on placebo, the patients were randomly assigned to either levoprotiline or maprotiline treatment for three weeks. In the next three weeks, responders were maintained on the same medication and non-responders were shifted to treatment with the complementary drug. After the initial three weeks' treatment, 31% of levoprotiline patients and 58% of maprotiline patients had responded. Both in the initial three-week treatment period and after shifting non-responders to the complementary drug, there were significant differences in favour of maprotiline. The comparison of properties of different outcome criteria in prediction analyses shows that the final score gives the best agreement with global evaluation; using the delta score (final minus baseline) or ratio score (final/baseline) as the outcome criterion may yield paradoxical results.