BACKGROUND: HIV increases the risk of tuberculosis directly, through immunosuppression, and indirectly, through onward transmission of Mycobacterium tuberculosis from the increased caseload. We assess the contribution of these two mechanisms by time since seroconversion to HIV. METHODS: The incidence of new pulmonary tuberculosis was estimated in a retrospective cohort study of South African gold miners over 14 years. HIV tests were done in random surveys in 1992-1993, and in clinics. One thousand nine hundred fifty HIV positive men with seroconversion intervals of less than 3 years were identified and linked to medical, demographic and occupational records. They were compared with men who were HIV-negative in a survey, with no later evidence of HIV. Analyses were censored when men were diagnosed with tuberculosis, died or left the mine. RESULTS: Tuberculosis incidence rose soon after HIV infection, reaching 1.4/100 person-years (95% confidence interval 1.1-1.9) within 2 years, and 10.0/100 person-years (95% confidence interval 6.5-15.5) at 10 or more years. By 11 years from seroconversion, nearly half the men had had tuberculosis. Among 5702 HIV-negative men, tuberculosis incidence was 0.48/100 person-years (95% confidence interval 0.33-0.70) in 1991-1993 and doubled over the period of the study (after adjusting for age). Age-adjusted model estimates suggest that half the increase in tuberculosis incidence by time since HIV infection was attributable to increasing incidence over calendar period--the indirect effect. CONCLUSION: For the first time, we have shown that the increase in tuberculosis risk by time since seroconversion reflects both direct effects of HIV increasing susceptibility, and indirect effects due to onward transmission. Innovative and sustained public health measures are needed to reduce Mycobacterium tuberculosis transmission.