Patients with advanced breast carcinoma and no prior chemotherapy were prospectively evaluated to assess the induction capabilities of cyclophosphamide, methotrexate and 5-fluorouracil (CMF), Adriamycin and vincristine (AV), and CMF plus prednisone (CMFP). The crossover responsiveness from CMF or CMFP to AV and of AV to CMF were also assessed. A disproportionate randomization led to 166 analyzable cases on AV, 79 on CMF were also assessed. A disproportionate randomization led to 166 analyzable cases on AV, 79 on CMF and 86 on CMFP induction. One hundred and twelve patients were evaluated on crossover. Induction response rates were similar with 56% on AV, 57% on CMF and 63% on CMFP. Crossover response rates ranged from 32% to 41%. CMFP and AV were superior to CMF in terms of response duration (P = 0.05), and CMFP was superior to either in terms of time to treatment failure (P = 0.04), and survival (P = 0.03). Treatment failures occurred in only the on-study organ sites of disease in 73% of the patients and did not appear to be related to the response achieved. CMF was associated with more thrombocytopenia than either AV or CMFP (P = 0.03). AV was associated with fewer infections than CMFP (P = 0.02), less diarrhea than CMFP (P = 0.04), more emesis than CMF (P = 0.02), and more neurologic toxicity than either CMF or CMFP (P less than 0.0001). There was also more emesis with CMF than with CMFP (P = 0.006). CMFP was associated with greater delivery of CMF than was the CMF regimen despite a similar day 1 leukocyte distribution. These results strongly suggest that CMF(P) and AV are clinically noncross-resistant regimens, that AV and CMF are essentially equivalently active induction regimens, and that CMFP is superior to CMF and AV.