OBJECTIVE: To determine the effectiveness of strategies to promote adherence to treatment for tuberculosis. IDENTIFICATION: Searches in Medline (1966 to August 1996), the Cochrane trials register (up to October 1996), and LILACS (Literatura Latinoamericana y del Caribe en Ciencias de la Salud) (1982 to September 1996); screening of references in articles on compliance and adherence; contact with experts in research on tuberculosis and adherence. INCLUSION CRITERIA: Randomised or pseudorandomised controlled trials of interventions to promote adherence with curative or preventive treatment for tuberculosis, with at least one measure of adherence. MAIN OUTCOME MEASURE: Relative risks and 95% confidence intervals for estimates of effect for categorical outcomes. RESULTS: Five trials met the inclusion criteria. The relative risk for tested reminder cards sent to patients who defaulted on treatment was 1.2 (95% confidence interval 1.1 to 1.4), for help given to patients by lay health workers 1.4 (1.1 to 1.8), for monetary incentives offered to patients 1.6 (1.3 to 2.0), for health education 1.2 (1.1 to 1.4), for a combination of a patient incentive and health education 2.4 (1.5 to 3.7) or 1.1 (1.0 to 1.2), and for intensive supervision of staff in tuberculosis clinics 1.2 (1.1 to 1.3). There were no completed trials of directly observed treatment. All of the interventions tested improved adherence. On current evidence it is unclear whether health education by itself leads to better adherence to treatment. CONCLUSIONS: Reliable evidence is available to show some specific strategies improve adherence to tuberculosis treatment, and these should be adopted in health systems, depending on their appropriateness to practice circumstances. Further innovations require testing to help find specific approaches that will be useful in low income countries. Randomised controlled trials evaluating the independent effects of directly observed treatment are awaited.