While directly observed treatment (DOT) has been recommended as the standard approach to tuberculosis control, empirical data on its feasibility and efficiency are still scarce. We conducted a controlled trial of DOT at 15 health care facilities at various levels of the government health care system in Thailand. A total of 836 patients diagnosed between August 1996 and October 1997 were randomly assigned to be treated either under DOT or self-supervised using monthly drug supplies (SS). Options for treatment supervisors were health staff, community members or members of the patients' families. Treatment outcomes were compared on the basis of cure, treatment-completion, default and death rates. In both study arms, treatment outcomes were improved compared to pre-study conditions. Cure and treatment-completion rates were significantly higher in the DOT cohort (76% and 84%) than in the SS group (67% and 76%). The benefits of DOT were more pronounced at district and provincial hospitals (DOT cure rate 81% vs. 69% in the SS group), while differences for patients treated at referral centres were non-significant (DOT cure rate 72% vs. 66% in the SS group). No significant differences in outcomes could be observed between patient groups receiving DOT under the various options for treatment supervisors. DOT appears especially suited for treatment at decentralized facilities. While a general focus on programme performance can improve outcomes, DOT provides significant additional benefits. If basic conditions are met, a DOT strategy can be tailored to country-specific conditions by exploring multiple observation options, without decreasing its effectiveness.