To evaluate the prognostic accuracy of N-staging by Ho's and the UICC/AJCC systems, 5020 patients with undifferentiated or poorly differentiated squamous cell carcinoma of the nasopharynx treated at the Queen Elizabeth Hospital, Hong Kong in the period 1976-1985 were analysed retrospectively. They were initially staged with Ho's system, but detailed records of nodal involvement allowed accurate retrospective restaging with the UICC/ AJCC (1988) system. Staging assessment depended almost entirely on physical examination; only 14% of patients had additional investigations with computed tomography. To evaluate the independent significance of different nodal parameters, T-stage adjusted analyses of the 4730 patients presenting without distant metastases were performed. Both N-staging systems showed a strongly significant overall correlation with distant failures and cancer-specific deaths. A significant trend was also shown for nodal failures in node-positive patients. Ho's system was superior in predicting distant failures, while the UICC/AJCC system was superior for nodal failures. However, even with due adjustment for level, the independent significance of nodal size, laterality and fixity could be demonstrated. After adjustment for UICC/AJCC N-stage, both level and fixity were also significant. Furthermore, when adjusted for all other meaningful parameters, there were no significant differences between ipsilateral and contralateral involvement, upper and mid-level extent, and nodal size < or = 3 cm or > 3- < or = 6 cm. N-staging can be further optimized by a newly proposed system incorporating fixity (movable versus fixed), level (upper-mid versus lower), size (greatest diameter < or = 6 cm versus > 6 cm), and laterality (unilateral versus bilateral) as staging criteria.