In a randomized, prospective, multicenter study with an observation period of 2.5 years, the differential prophylactic efficacy of lithium versus carbamazepine was compared in 171 patients fulfilling DSM-IV criteria for bipolar disorder. Serum drug levels were 0.6+/-0.1 mmol/L for lithium and 6.1+/-1.3 microg/mL for carbamazepine. Patients were subdivided into a classical subgroup (bipolar I patients without mood-incongruent delusions and without comorbidity, N = 67) and a nonclassical subgroup including all other patients (N = 104). Classical bipolar patients had a lower rehospitalization rate with lithium than with carbamazepine prophylaxis (p = 0.005). For the nonclassical group, a trend in favor of carbamazepine was found. In the lithium group, there was a positive association between hospitalization rate and number of nonclassical features (bipolar II/not otherwise specified, mood-incongruent delusions, comorbidity; p = 0.035). For carbamazepine, this association was negative (p = 0.033). Analyses including mixed states as an additional nonclassical feature confirmed the results. In conclusion, lithium seems to be superior to carbamazepine in treating classical bipolar cases. Patients with nonclassical features might profit more from prophylaxis with carbamazepine, which seems to have a broader spectrum of activity.