BACKGROUND A retrospective analysis of treatment outcomes in patients with nasopharyngeal carcinoma (NPC) was performed in which the newly revised 1997 American Joint Committee on Cancer (AJCC) stage classification was applied and compared with the 1988 AJCC and Ho stage classifications, with emphasis on the predictive value of different staging systems in determining failure patterns in NPC. METHODS Three hundred and twenty-four patients with newly diagnosed NPC treated between September 1989 and August 1991 and originally staged according to Ho stage classification were re-staged according to the 1988 and 1997 AJCC stage classifications. In addition to stage grouping, patients were also classified into the following prognostic categories to study the failure patterns: early disease group (T1-2N0-1), advanced local disease group (T3-4N0-1), advanced nodal disease group (T1-2N2-3), and advanced locoregional disease group (T3-4N2-3). The overall survival (OAS), relapse-free survival (RFS), local relapse-free survival, nodal relapse-free survival, and distant metastases-free survival were compared among different stage groups and prognostic categories in the three staging systems. RESULTS In the new AJCC system, the percentages of patients with Stage I, II, III, and IV disease were 15.1%, 31.5%, 28.1%, and 25.3%, respectively, whereas most patients were classified as having Stage IV disease (65.7%) in the old AJCC system and Stage II or III disease (74.1%) in the Ho system. The 5 year OAS rates in the 1997 AJCC Stage I, II, III, and IV disease were 97.7%, 78.7%, 79.5%, and 61.4%, respectively. The corresponding 5 year RFS rates were 95.7%, 64.7%, 54.5%, and 41.1%. Using the 1997 AJCC system to define the four prognostic categories, the early disease group had the lowest incidence of relapse (27.6%) and death (18.4%), whereas the advanced locoregional disease group had the highest incidence of relapse (61.4%) and death (43.2%). Both the advanced local disease group and the advanced nodal disease group had similar rates of relapse (46.7% vs. 47.2%), but local relapse was the major cause of failure in the former group (61.8%), whereas distant metastases was the major cause in the latter group (44%). CONCLUSIONS Using the 1997 AJCC staging system, the authors observed a better distribution of patient numbers as well as segregation of survival curves among different stage groups. Moreover, prognostic categories with distinct prognosis and failure patterns were definable by the new system, which has important implications in selecting appropriate patient treatment strategies.