OBJECTIVE: To evaluate the diagnostic performance of manual MGIT (MMGIT) compared to the gold standard, Lowenstein-Jensen (LJ), in the diagnosis of pulmonary tuberculosis (TB) in a high-burden setting. METHODS: Individuals with suspected TB enrolled in parallel diagnostic trials during 2007-2011 were included. Two samples were obtained from each patient and inoculated into MMGIT and LJ medium. Diagnostic tests were performed, and the incremental yield of a second test and time to detection (TTD) were calculated. Analyses were performed per patient and per sample. Gold standard was based on LJ culture. RESULTS: In the per patient and per sample analysis, we evaluated 1436 patients and 4142 samples. The sensitivity and specificity for smear and MMGIT per sample were respectively 89.9%/92.2% and 97.1%/98.9%. Contamination was observed in 1.4% of samples on MMGIT. The mean TTD (days) was 11.8 for MMGIT and 22.9 for LJ. The sensitivity and specificity for smear and MMGIT per patient were respectively 89.9% and 92.2% and 97.1% and 98.3%. A second MMGIT culture had an incremental yield of 1.6%. CONCLUSIONS: MMGIT has high sensitivity and specificity, regardless of smear result, with a 50% reduction in TTD compared to LJ. These features make MMGIT an acceptable TB diagnostic method for use in resource-limited settings.