[en] Purpose/Objective: We sought to evaluate the prognostic efficacy of a new staging system for Head and Neck (H and N) cancers that was derived by recursive partitioning analysis (RPA). Materials and Methods: RPA is a statistical technique that can separate heterogeneous populations into homogeneous subgroups. It requires no a priori knowledge of potential prognostic factors and can create groupings based on different outcome measures, such as local-regional relapse or survival. We analysed the outcome of 2,105 patients, who had squamous cell carcinomas of the H and N region that were irradiated as part of four RTOG protocols. RPA created two different sets of subgroups (i.e., defined stages) based upon (1) survival and (2) local-regional relapse. These RPA determined stages were compared to AJC defined stages. Results: RPA created six stages of disease when survival was the measure of outcome and five stages when local-regional relapse was used, as compared to four stages in the AJC system. Although the assignment of tumors to stage groupings in the two systems did correlate, there was substantial disparity in assignment between the systems. This suggests that RPA recognizes factors that affect outcome which are not recognized by the AJC system. Conclusion: RPA derived staging provides an alternative way to classify H and N tumors. It can (1) disclose the presence of factors currently not recognized by the AJC system, (2) provide an additional basis for selection of patient specific therapy, and (3) potentially be used to improve the AJC system in the future