BACKGROUND:: Tuberculosis-HIV (TB-HIV) coinfection remains an important cause of mortality in antiretroviral therapy (ART) programs. In a cohort of TB-HIV-coinfected patients starting ART, we examined the incidence and predictors of early mortality. METHODS:: Consecutive TB-HIV-coinfected patients eligible for ART were enrolled in a cohort study at the Mulago National Tuberculosis and Leprosy Program clinic in Kampala, Uganda. Predictors of mortality were assessed using Cox proportional hazards analysis. RESULTS:: Three hundred and two patients [median CD4 count 53 cells/μL (interquartile range, 20-134)] were enrolled. Fifty-three patients died, 36 (68%) of these died within the first 6 months of TB diagnosis. Male sex [hazard (HR): 2.19; 95% confidence interval (CI): 1.19 to 4.03; P = 0.011], anergy to tuberculin skin test [HR: 2.59 (1.10 to 6.12); P = 0.030], a positive serum cryptococcal antigen result at enrollment (HR: 4.27; 95% CI: 1.50 to 12.13; P = 0.006) and no ART use (HR: 4.63; 95% CI: 2. 37 to 9.03; P textless 0.001) were independent predictors of mortality by multivariate analysis. Six (10%) patients with TB immune reconstitution inflammatory syndrome died, and in most, an alternative contributing cause of death was identified. CONCLUSIONS:: Mortality among these TB-HIV-coinfected patients was high particularly when presenting with advanced HIV disease and not starting ART, reinforcing the need for timely and joint treatment for both infections. Screening for a concomitant cryptococcal infection and antifungal treatment for patients with cryptococcal antigenemia may further improve clinical outcome.