We have compared the rate of recovery of mycobacteria with the MB-Check culture system (liquid phase) and the Löwenstein-Jensen (LJ) medium in 2,907 clinical specimens obtained from 830 patients submitted for mycobacterial culture during 1-year period. Direct smear examination was carried out by auramine-rhodamine staining. All primary isolates from the culture media were confirmed by Ziehl-Neelsen staining and identified by acridinium-ester-labeled DNA probes specific for Mycobaterium tuberculosis complex. A total of 214 isolates were of the M. tuberculosis complex (88 patients) and 54 of "potentially pathogenic environmental mycobacteria" (45 patients). A total of 117 (54.7%) samples were smear-positive and the remaining 97 (45.3%) were smear-negative. There was a significant difference in the percentage of positive cultures obtained by the MB-Check method (99.1%) as compared with the LJ medium (73.8%) (P < 0.05). This difference, however, occurred almost exclusively at the expense of the 97 smear-negative samples (positive cultures 97.95% by the MB-Check method vs. 42.3% by the LJ culture, P < 0.05). The number of patients diagnosed of tuberculosis by the MB-Check was significantly higher as compared with LJ medium (88 [100%] vs. 77 [87.5%], P < 0.05). In 11 (12.5%) patients, the diagnosis was only established by the MB-Check system. In smear-positive samples, the mean (+/-SD) detection time for M. tuberculosis complex was 14.8 +/- 8 days with MB-Check and 19.9 +/- 7 days with LJ medium. The corresponding figures in smear-negative samples were 22.8 +/- 3 days and 27.8 +/- 6 days, respectively. DNA probes directly applied to MB-Check liquid medium showed a sensitivity of 98.8% and specificity of 100%. These results indicate that the MB-Check system is more efficient for the recovery of mycobacteria than LJ medium.