Identification of important prognostic factors with respect to the patient's initial response to therapy was made from a set of 25 covariates available on 281 patients with advanced breast cancer. Since the patients studied were all participants in a randomized clinical trial that involved three different treatment regimens (repeated weekly treatment of a combination of cyclophosphamide, methotrexate, 5-fluorouracil, vincristine, and prednisone; intermittent treatment of the same preceding five drugs given in 5-day courses every 4 weeks; or treatment every 3 weeks with adriamycin as a single agent), the effect of these treatments on the selection of important covariates was assessed. Although some evidence indicated that the set of important covariates differed by treatment, the differences were not strong enough to be of practical importance. Non-Caucasian patients did poorly on all regimens with a response rate of only 31% compared to 62% for Caucasian patients. The covariates of major prognostic importance for Caucasians were: disease-free interval, liver involvement, and performance status. Ambulatory patients with long disease-free intervals and no liver involvement had the best prognosis. After adjustments were made for these three covariates, the remaining covariates (such as menopausal status, bone involvement, number of metastatic sites, and duration of metastatic disease) were not significantly related to response. As reported earlier, the treatment effect was significant, even after adjustments were made for the important covariates.