Tuberculosis (TB) remains a leading cause of death among HIV-infected adults, in part because of delayed diagnosis and therefore delayed initiation of treatment. Recently, the Gene-Xpert platform, a rapid, PCR-based diagnostic platform, has been validated for the diagnosis of TB with sputum. We have evaluated the Xpert MTB/RIF assay for the diagnosis of Mycobacterium tuberculosis bacteremia and investigated its impact on clinical outcomes. Consecutive HIV-infected adults with fever and cough presenting to Queen Elizabeth Central Hospital, Blantyre, Malawi, were recruited and followed up for 2 months. At presentation, three sputum samples were examined by smear, culture, and Xpert MTB/RIF assay for the presence of M. tuberculosis and blood was drawn for PCR with Xpert, for mycobacterial culture (Myco/F Lytic), and for aerobic culture. One hundred four patients were recruited, and 44 (43%) were sputum culture positive for M. tuberculosis. Ten were Xpert blood positive, for a sensitivity of 21% and a specificity of 100%. The 2-week mortality rate was significantly higher among patients who were Xpert blood positive than among those who were negative (40% versus 3%; multivariate odds ratio [OR] for death if positive, 44; 95% confidence interval [CI], 3 to 662). This effect persisted on assessment of the mortality rate at 2 months (40% versus 11%; OR, 5.6; 95% CI, 1.3 to 24.6). When screening uncomplicated patients presenting with a productive cough for pulmonary TB, Xpert blood offers no diagnostic advantage over sputum testing. Despite this, Xpert blood positivity is highly predictive of early death and this test rapidly identifies a group of patients in urgent need of initiation of treatment.