OBJECTIVE: To evaluate the association between a positive serum cryptococcal antigen (CRAG) test at baseline and mortality during the first 12 weeks on antiretroviral therapy (ART). Cryptococcal meningitis is a leading cause of HIV-related mortality in Africa, but current guidelines do not advocate CRAG testing as a screening tool. METHODS: Between May 2003 and December 2004, we enrolled HIV-1 infected individuals into a study of ART monitoring in rural Uganda. CRAG testing was conducted retrospectively on stored pre-ART serum samples of participants whose baseline CD4 cell count was <100 cells/mul and who were without symptoms suggestive of disseminated cryptococcal disease at enrolment. RESULTS: Of 377 participants, 5.8% had serum CRAG titre >/=1:2. Of these, 23% died during follow-up. Controlling for CD4 cell count, HIV-1 viral load, anaemia, active tuberculosis and body mass index, relative risk of death during follow-up among those with asymptomatic cryptococcal antigenemia at baseline was 6.6 [95% confidence interval (CI) 1.86-23.61, P = 0.0036]. The population attributable risk for mortality associated with a positive CRAG at baseline was 18% (CI 2-33%), similar to that associated with active tuberculosis (19%, CI 1-36%). CONCLUSION: Asymptomatic cryptococcal antigenemia independently predicts death during the first 12 weeks of ART among individuals with advanced HIV disease in rural Uganda. Routine screening and provision of azole antifungal therapy prior to or simultaneous with the start of ART should be evaluated for the potential to prevent mortality in this population.