1083 patients with late-stage Trypanosoma brucei gambiense sleeping sickness were treated with melarsoprol in Nioki hospital, Zaire, between 1983 and 1990. Sixty-two (5·7%) died during treatment. Of the 1021 patients who survived the treatment, 63 (6·2%) subsequently relapsed, 58 (92%) of whom were diagnosed within 2 years of melarsoprol treatment. There was no evidence of an increase in the frequency of treatment failures during the study period, and the rate of relapses that we documented is comparable to that reported from Zaire more than 30 years ago. Relapses were more frequent among patients who had trypanosomes seen in the cerebrospinal fluid (CSF) at the time of the initial diagnosis (odds ratio [OR] = 2·76, 95% confidence interval [CI] = 1·65–4·63, P = 0·0001). Male patients had twice as many relapses as females (OR = 2·00, 95% CI = 1·19–3·36, P = 0·009), which was partly explained by males having trypanosomes in the CSF more often than females. There were important geographical variations in the frequency of relapses within the territory of the Nioki rural health zone, suggesting that the circulation of trypanosomes was geographically limited. Prednisolone treatment did not increase the risk of treatment failure, nor did decreasing the total dose of melarsoprol from 12 to 9 injections for patients with ⩾100 white blood cells/mm3 of CSF. Since patients with trypanosomes in the CSF are also those who are at the highest risk of melarsoprolinduced encephalopathy, more aggressive treatment regimens cannot be recommended. Indeed our data suggest that there may be a threshold above which further increasing the total dosage of melarsoprol will not reduce the risk of relapse.