OBJECTIVE To establish the optimum fractionation for high-dose-rate (HDR) endoesophageal brachytherapy (EBT) for early stage esophageal cancer from retrospective data of patients treated with different HDR schedules following external beam irradiation (EBI). METHODS The study population consisted of 35 consecutive early stage esophageal cancer patients who received EBI to the mediastinum, plus EBT, between May 1992 and November 1995 at the Hiroshima University Medical Center and Hiroshima City Hospital. All patients were treated with EBI, with doses ranging from 50 to 61 Gy. The spinal cord was spared after 44-45 Gy. HDR EBT was performed using a double-balloon applicator in conjunction with an Ir-192 remote afterloading system. One group of 10 patients was given a weekly endoesophageal boost of 4 or 5 Gy at a distance of 5 mm from the applicator surface over a period of 1-2 weeks. Another group of 25 patients was treated with 4 or 5 endoesophageal boosts with a fraction dose of either 2.5 or 2 Gy for 1 week. The linear quadratic (LQ) formula was used to calculate the biologically effective dose (BED) for tumor (Gy10) and esophageal mucosa (Gy3); Gy10 means alpha/beta equals 10 Gy, and Gy3 means alpha/beta equals 3 Gy. The Kaplan-Meier method was used to calculate the local control and late complication rates, while the Cox-Mantel test was used to evaluate statistical significance (p < 0.01). RESULTS Nine (26%) of the 35 patients recurred locally and 7 (20%) had late complications (esophageal ulcer grade by RTOG/EORTC criteria > 1). The 5-year overall survival, local control, and late complication rates were 38%, 57%, and 26%, respectively. The probability of local recurrence was not affected by the treatment parameters. Results from the LQ formula significantly correlate with data on late complications. A BED > 134 Gy3 and a fraction number = < 3 were associated with late complications (grade > 1). BED analysis showed that the fractionation dose should be decreased to 2.5 or 2.0 Gy at a distance of 5 mm from the applicator surface, and the number of doses increased to 4 or 5, respectively, to yield a satisfactory BED (< 134 Gy3). CONCLUSIONS A significant reduction in endoesophageal brachytherapy dose per fraction is necessary to reduce late complications. Our current treatment protocol for early-stage esophageal cancer consists of EBI of 60 Gy followed by 4 EBT doses at a fraction dose of 2.5 Gy applied over 1 week.