The reliability of the Mycobacteria Growth Indicator Tube (MGIT) 960 system for rapid detection of mycobacteria in clinical specimens was evaluated and compared to the radiometric method (BACTEC 460TB) and to mycobacterial culture on Lowenstein-Jensen (LJ) medium. Clinical specimens (n = 590) were tested without selection. A total of 121 (20.5%) isolates of mycobacteria were recovered; 98 (81.0%) of them were recovered with the BACTEC 460TB system, 86 (71.1%) were recovered with the BACTEC MGIT 960 system, and 55 (45.5%) were recovered with LJ medium (MGIT 960 versus BACTEC 640TB, p >0.05; MGIT 960 or BACTEC 460TB versus LJ, p <0.001). The mean time to detection (TTD) was 18 da for BACTEC 460 TB, and 13 da for BACTEC MGIT 960. The mean time to detection in each system, based upon data where both systems were culture positive, was significantly different (16.6 da for BACTEC 460TB and 13 da for BACTEC MGIT 960, p<0.001). The contamination rate of the BACTEC MGIT 960 system was 13.2%, which was intermediate between the BACTEC 460TB system (11.7%) and the LJ medium (14.7%). These data indicate that the fully automated MGIT 960 system is an accurate, non-radiometric alternative to the BACTEC 460TB method for rapid detection of mycobacteria in a clinical setting.