A retrospective chart review was conducted of women with stage III breast cancer seen at the Princess Margaret Hospital between January 1977 and December 1980. Three hundred and sixty-nine patients were available for analysis. These cases were evaluated to determine the prognostic factors of patients presenting with this stage of the disease using a recursive partitioning technique, RECPAM, and a Cox regression model. A non-mathematical description of RECPAM is presented and the advantages of RECPAM over Cox analysis are discussed. The results identify primary tumour size, axillary node involvement, internal mammary node involvement, and estrogen receptor status as the most important prognostic variables. RECPAM identified 3 prognostic groups and simultaneously provided rules based on the prognostic variables to assign patients to poor, intermediate, or good prognosis categories. Patients with estrogen receptor negative tumours, or those with axillary node involvement, primary tumours greater than 5 cm, and serum alkaline phosphatase greater than 60 IU/L, or those with internal mammary node involvement, no skin changes, and serum alkaline phosphatase greater than 60 IU/L, define a group with a poor prognosis. Patients with estrogen receptor positive tumours, no axillary node involvements, and primary tumours greater than 5 cm, or estrogen receptor positive tumours, axillary node involvement, primary tumours greater than 5 cm, but serum alkaline phosphatase less than or equal to 60 U/L, have an intermediate prognosis. The good prognosis group consists of those patients with estrogen receptor positive tumours who have either skin changes or primary tumours less than or equal to 5 cm. The effect of loco-regional and systemic therapy was assessed and there was no association between treatment assignment and prognostic group. On the basis of this RECPAM analysis, we have defined important prognostic variables to be used in the design of clinical trials, and three major patient subgroups which can be used in routine oncologic practice as a guide to patient management.