OBJECTIVE To study the treatment outcome in patients with locally recurrent nasopharyngeal carcinoma (NPC) and to explore whether a combination of high-dose-rate (HDR) intracavitary brachytherapy and external beam radiation therapy (ERT) could improve the therapeutic ratio. METHODS Ninety-one patients with nonmetastatic locally recurrent NPC who were treated with curative intent during the years 1990-1999 were retrospectively analyzed. Eighty-two patients had histologically proven carcinoma. The remaining 9 had clinical and imaging features suggestive of local recurrence. The Ho's T-stage distribution at recurrence (rT) was as follows: rT1-37, rT2-14, rT3-40. Total equivalent dose (TED) was calculated by the linear-quadratic formula without a time factor correction. For those treated by combined-modality treatment (CMT), the TED was taken as the summation of the equivalent dose by ERT and the absolute dose delivered to floor of the sphenoid by brachytherapy. Eight patients were treated solely with brachytherapy, all receiving 24-45 Gy in 3-10 sessions. Forty-one patients were treated with ERT alone receiving a median TED of 57.3 Gy (range, 49.8-62.5 Gy). Forty-two patients were treated by CMT with a median equivalent dose of 50 Gy (range, 40-60 Gy) given by ERT and 14.8 Gy by brachytherapy (range, 3-29.6 Gy). Multivariate analyses were performed using the Cox regression proportional hazards model. RESULTS The 5-year actuarial overall survival rate, disease specific survival rate and local failure-free survival (LFFS) rate for the whole group were 30%, 33. 3% and 37.8%, respectively. The 3-year LFFS rates of rT1, rT2, and rT3 diseases were 64%, 61.5%, and 18.4%, respectively (p = 0.001). Of the 8 patients treated with brachytherapy alone, 4 failed locally. Further analyses were concentrated on the ERT (41 patients) and CMT (42 patients) groups. The 3-year LFFS rates of rT1, rT2, and rT3 diseases were 66.7%, 66.7%, and 18.4%, respectively (p = 0.0008). Better local control for patients who received a TED of 60 Gy or greater was shown. The corresponding 3-year LFFS rates were 29% and 60% (p = 0.0004). Subgroup analysis on the ERT and CMT groups showed a 3-year LFFS rate of 33.5% and 57% (p = 0.003). ERT group had an excess of patients with rT3 disease. Further analysis was performed on the rT1-2 patients showing a trend toward improvement in local control in favor of the CMT group (3-year LFFS rates: CMT, 71.7%; ERT, 54%; p = 0.13). Multivariate analyses showed that rT stage (p = 0.002) and TED (p = 0.01; HR, 0.93; 95% confidence interval, 0.88-0. 98) remained significant. The 5-year major and central nervous system (CNS) complication-free rates were 26.7% and 47.8%. The following factors were found to be significant on univariate analyses for both complications in the ERT and CMT groups: (1) Modality of treatment: more complications with ERT group; and (2) rT stage. Multivariate analyses showed that the rT stage was significant for predicting the occurrence of major (p = 0.004) and CNS complications (p = 0.04). CONCLUSIONS For rT1-2 local recurrences, CMT with at least 60 Gy TED is recommended. The high incidence of major late complications is of serious concern. Ways of improving the local control of Ho's rT3 disease and reducing the risk of late complications should be explored.