The reliability of the BDProbeTec MTB Test (Becton Dickinson, Sparks, Md.) for direct detection of Mycobacterium tuberculosis in respiratory specimens was evaluated by comparing results to those of conventional mycobacterial culture, with the BACTEC TB 460 and Middlebrook 7H11 biplates. Patients known to have tuberculosis were excluded from analysis. Of 523 specimens from 277 patients, 53 grew a mycobacterium: 24 specimens of M. tuberculosis and 29 specimens of nontuberculous mycobacteria. After initial testing, 42 specimens were positive by the BDProbeTec, for overall sensitivity, specificity, and positive and negative predictive values of 95.8, 96. 2, 54.8, and 99.8%, respectively. After resolution of discrepancies, 28 specimens were positive by the BDProbeTec, for overall sensitivity, specificity, and positive and negative predictive values of 100, 99.2, 85.7, and 100%, respectively. These same values were 100, 80.8, 93.4, and 100%, respectively, for smear-positive samples and 100, 99.4, 75.0, and 100%, respectively, for smear-negative specimens.