Six hundred twenty-four women with metastatic breast cancer were entered on Eastern Cooperative Oncology Group (ECOG) study EST 2181. Patients were treated with mitolactol, doxorubicin, vincristine (DAV), tamoxifen, and fluoxymesterone (DAVTH). Nine patients were canceled, and 114 were ineligible (half because of concomitant diseases). Among the 501 eligible patients, the overall response rate was 54% (14% complete response and 5% not assessable). The median time to treatment failure (TTF) was 9.0 months, and the median survival was 20.9 months. Multivariate models were fit on a randomly chosen half of the eligible cases and then verified on the other half. About half of the variables that were significant in the models remained significant in the verification data set. In the verification data set the variables that remained significantly associated with lower probability of response were three or more organ sites of disease and lack of nodal metastases; the variables associated with a significantly shorter TTF were liver metastases, estrogen receptor (ER)-negativity, and prior adjuvant therapy. The variables associated with significantly shorter survival were liver metastases, ER negativity, three or more organ sites of disease, and prior adjuvant chemotherapy. None of the variables in the data set had a significant influence on toxicity. The 125 patients aged over 65 years did not have worse toxicity or worse prognosis than younger patients. Ineligible patients had significantly less response but virtually identical TTF curves, survival curves, and toxicities. Therefore, patient discriminants are of paramount importance in predicting the outcome of treatment. Many of the current criteria for eligibility for entry on study may not be justified.