Tuberculosis has reemerged as an important public health problem, and the frequency of drug resistance is increasing. A major reason for the development of resistant infections and relapse is poor compliance with medical regimens. In Tarrant County, Texas, we initiated a program of universal directly observed treatment for tuberculosis. We report the effect of the program on the rates of primary and acquired drug resistance and relapse among patients with tuberculosis. We collected information on all patients with positive cultures for Mycobacterium tuberculosis in Tarrant County from January 1, 1980, through December 31, 1992. Through October 1986, patients received a traditional, unsupervised drug regimen. Beginning in November 1986, nearly all patients received therapy under direct observation by health care personnel. A total of 407 episodes in which patients received traditional treatment for tuberculosis (January 1980 through October 1986) were compared with 581 episodes in which therapy was directly observed (November 1986 through December 1992). Despite higher rates of intravenous drug use and homelessness and an increasing rate of tuberculosis during this 13-year period, the frequency of primary drug resistance decreased from 13.0 percent to 6.7 percent (P<0.001) after the institution of direct observation of therapy, and the frequency of acquired resistance declined from 14.0 percent to 2.1 percent (P<0.001). The relapse rate decreased from 20.9 percent to 5.5 percent (P<0.001), and the number of relapses with multidrug-resistant organisms decreased from 25 to 5 (P<0.001). The administration of therapy for M. tuberculosis infection under direct observation leads to significant reductions in the frequency of primary drug resistance, acquired drug resistance, and relapse.