The primary objective of this multicenter, open-label, randomized, parallel, phase II selection trial was to compare the objective tumor response to biweekly (every 2 weeks) gemcitabine/paclitaxel, gemcitabine/carboplatin, and gemcitabine/cisplatin as first-line treatment for metastatic breast cancer. Eligible patients with stage IV disease who relapsed after anthracycline failure were randomly assigned in a 1:1:1 ratio to gemcitabine (2,500 mg/m(2)) plus paclitaxel 150 mg/m(2) (n = 49); plus carboplatin, area under the curve = 2.5 mg/mL x min (n = 47); or plus cisplatin 50 mg/m(2) (n = 51). Study therapy continued up until a maximum of 8 cycles and follow-up continued for 24 months. All patients were analyzed for efficacy and one patient was excluded from the safety analyses. The objective response was 26.5% [95% confidence interval (CI) 14.9-41.1] for gemcitabine/paclitaxel, 17.0% (95% CI 7.6-30.8) for gemcitabine/carboplatin, and 15.7% (95% CI 7.0-28.6) for gemcitabine/cisplatin. The adjusted odds ratio for tumor response was 0.33 (95% CI 0.10-1.06), P = 0.063 for gemcitabine/carboplatin versus gemcitabine/paclitaxel; 0.26 (95% CI 0.08-0.86), P = 0.027 for gemcitabine/cisplatin versus gemcitabine/paclitaxel; and 0.77 (95% CI 0.24-2.52), P = 0.671 for gemcitabine/cisplatin versus gemcitabine/carboplatin. There were no significant differences in overall survival or progression-free survival (P > 0.05). Grade 3 or 4 drug-related adverse events varied between groups and the majority of deaths (94.9%; 74/78) were related to disease progression. The gemcitabine-based treatments had comparable activity and tolerability. Similar survival characteristics and different toxicity profiles suggested that gemcitabine-platinum may be evaluated further in patients after anthracycline failure.