Several putative prognostic factors have been identified in node-positive breast cancer patients, but their importance needs to be clarified in a uniformly treated population. The objectives of this investigation were: 1) to describe the characteristics of a uniformly treated node-positive data base; 2) to use proportional hazards (Cox) and recursive partitioning and amalgamation (RPA) multivariate models to assess the importance of potential prognostic factors for disease-free and for overall survival; and 3) to define prognostic groups with different disease-free survival and survival outcomes with RPA. A data base of 768 node-positive patients enrolled on 1-year adjuvant CMFVP arms of four SWOG trials was formed. Variables were number of positive nodes, age, age at menopause, menopausal status, ER status, ER and PgR levels (for RPA only), tumor size, race, breast cancer in mother, and obesity index. Independent predictors of both disease-free and overall survival in the Cox models were: number of positive nodes (4-6 worse than 1-3, and better than greater than 6); the age/menopause category (age greater than or equal to 35/premenopausal better than age less than 35/premenopausal and better than postmenopausal); and ER status (patients on ER-negative study worse than others). The RPA for disease-free survival defined four subgroups based on nodes, menopausal status, tumor size, and age at menopause (5-year recurrence-free rates = 73%, 52%, 38%, and 15%). The RPA for survival found four prognostic groups, defined only by the number of positive nodes and ER and PgR levels (5-year survivals = 91%, 72%, 56%, and 37%). Both RPAs suggested interesting refinements of the results of the Cox models. In the RPA for disease-free survival, best node cutoffs differed by menopausal status, tumor size was important only in postmenopausal patients with few positive nodes, and age at menopause emerged as an independent predictor of recurrence potential. And, the RPA for survival showed that node cutoffs differed according to ER level. Thus, these analyses underscore the value of simple, clinically available prognostic factors and suggest the possible need to reconsider the definition of good and poor risk patient groups in future adjuvant trial design.