Between 1979 and 1984, 185 fully evaluable patients with stage III or IV epithelial type ovarian cancer and suboptimal surgical resections were randomly assigned to treatment with doxorubicin + cyclophosphamide + BCG (DC + BCG) vs doxorubicin + cyclophosphamide + cisplatin (DCP) vs. doxorubicin + cyclophosphamide + cisplatin + BCG (DCP + BCG). Patients with measurable disease (119) were analyzed separately from those with nonmeasurable disease (66). In measurable disease patients the overall clinical complete plus partial response rates for DC + BCG, DCP, and DCP + BCG-treated patients were 36, 57, and 59%, respectively. Although there were no significant patient characteristic differences between the DCP and DCP + BCG treatment groups, the addition of cisplatin to the DC + BCG regimen resulted in significantly prolonged response (P less than 0.03) and survival (P less than 0.002) durations. To the contrary, the addition of BCG to the DCP regimen did not improve objective response rates or response or survival durations. For patients with nonmeasurable, suboptimal disease there were no significant differences between the three treatments with respect to response or survival parameters; however, patients in this disease category fared generally better than those with clinically measurable disease. We conclude that cisplatin adds significantly to the efficacy of DC + BCG, but BCG does not add to the efficacy of DCP in patients with measurable, stage III or IV disease.