The series consisted of 759 patients with metastatic breast cancer entered into randomized clinical trials at the Curie Institute. Twenty factors were found to be significant by univariate analysis. The current report gives a detailed analysis of prognostic factors using a new method of multivariate analysis for survival data, the recursive partition. This method was based on the construction of a regression scheme consisting of eight variables and four prognostic groups. A test sample procedure was used to validate our results and a regression scheme with three variables was constructed (LDH, adjuvant chemotherapy and the Karnofsky scale). The results were compared to the stepwise Cox regression.