The Roche Cobas Amplicor MTB assay, recently replaced by the Roche Cobas TaqMan MTB assay, was one of the first commercially available assays for detection of the Mycobacterium tuberculosis complex based on nucleic acid amplification. We reported previously on the limited specificity of the Cobas Amplicor MTB assay, in particular for positive samples with an optical density at 660 nm (OD660) of <2.0. Using a selected set of respiratory samples, which were scored as false positive by the Cobas Amplicor test, we demonstrate here that the specificity of the Cobas TaqMan assay is significantly improved. In addition, our study of a set of 133 clinical samples revealed that the Cobas TaqMan MTB assay showed significantly less PCR inhibition than the Cobas Amplicor test. An overall concordance of 98.2% was observed between the two assays. In a subsequent prospective study, we evaluated the performance of the Roche Cobas TaqMan MTB assay on 1,143 clinical specimens, including respiratory (n = 838) and nonrespiratory (n = 305) specimens. Using culture as the gold standard, we found a sensitivity of 88.4% and a specificity of 98.8% for the 838 respiratory specimens, compared to a sensitivity of 63.6% and a specificity of 94.6% for the 305 nonrespiratory specimens. We conclude that the Cobas TaqMan MTB assay is a significantly improved tool for the direct detection of M. tuberculosis DNA in clinical specimens.