BACTEC and conventional methods of antimicrobial susceptibility testing were compared with the use of artificial mixtures of 1% resistant and 99% susceptible Mycobacterium tuberculosis strains. Inocula for the assays were prepared on the basis of radiometric readings. A total of 40 resistant strains were tested: 18 were resistant to isoniazid, 16 to rifampin, 5 to streptomycin, and 1 to ethambutol. The BACTEC method detected 27 of 39 strains at the greater than 0.5% resistance level, whereas the conventional plate method detected only 8 of 40. In addition, the results of the BACTEC assay were closer to the expected 1% resistance level (0.7%) than were the data obtained with the proportional plate method (0.3%; P = 0.001). This difference was most striking with the isoniazid-resistant strains. The precision of the assay methods was quite different (coefficients of variation, 26% for BACTEC and 54% for plates; P = 0.0002). Streptomycin-resistant strains had the highest variability in both assay methods. A 1-day delay in calculating the BACTEC data after the control vial reading was equal to or greater than 30 resulted in a 25% reduction in the calculated level of resistance (0.7 to 0.53). By adhering exactly to the recommended procedures for the BACTEC test method and by using log-phase inocula, this method shows better precision and accuracy at the 1% resistance level than does the proportional plate method.