Artibonite Valley, a rural area in Haiti. To evaluate a tuberculosis control program in rural Haiti and to compare two strategies for treatment implemented in two areas that were not chosen at random: treatment delivered at the patients' homes observed by former tuberculosis patients (DOT), and non observed treatment (non-DOT). Retrospective analysis of the clinical records of adult patients diagnosed with tuberculosis at H pital Albert Schweitzer in Deschapelles, Haiti, during 1994-1995. There were 143 patients in the non-DOT group and 138 patients in the DOT group. The results of treatment were significantly different: in the non-DOT group 29% defaulted, 12% died and 58% had a successful outcome; in the DOT group 7% defaulted (P < 0.01), 4% died (P = 0.01) and 87% had a successful outcome (P < 0.01). These differences are also significant when considering only human immunodeficiency virus (HIV) infected patients (defaulted P < 0.01; died P = 0.09; successful outcome P < 0.01). Delivering treatment in patients' homes with direct observation by former tuberculosis patients can achieve good results, even in an area of extreme poverty and high rates of HIV infection. In this population the number of patients who are able to complete their treatment without observed administration is far from optimal.