Since response to chemotherapy is a major determinant of survival in metastatic breast cancer, the purpose of our study was to analyse the predictive factors of response. 1426 patients enrolled into eight consecutive randomised trials of anthracycline-based first-line chemotherapy in metastatic breast cancer, between 1977 and 1992, were analysed. A forward stepwise logistic regression analysis was used. The objective response rate (ORR) to chemotherapy in the total population was 63.6% (95% confidence interval (CI): 61.5-67.7). The complete response rate was 17.5%. Multivariate analysis defined adjuvant chemotherapy, lactate dehydrogenase (LDH), Karnofsky index (KI), and pleural and lung metastases to be the five main variables correlated with ORR. A predictive score was calculated using the coefficient of these five variables, The score was established as follows: -1.32+0.54 (if prior adjuvant chemotherapy) +0.80 (low KI) +0.75 (raised LDH) +0.49 (lung metastases) +0.51 (pleural metastases). A low score (less than -0.78) was associated with an ORR greater than 70.0%, representing 41.2% of our population. An intermediate score (between -0.78 and 0) was associated with an ORR of 50 to 70%, representing 37.5% of our population and a positive score was associated with an ORR of less than 50%, representing 21.3% of our population. This score can be used to predict objective response rates to first-line anthracycline-based chemotherapy. This method now needs to be evaluated prospectively in phase II trials. Identification of various risk groups may also be useful for interpretation and design of clinical trials.