A multicenter study was carried out to evaluate the performance of a new commercial automated system in comparison with that of the reference agar dilution method. Ten clinical microbiology laboratories tested a collection of 61 strains of gram-negative bacilli (49 Enterobacteriaceae and 12 Pseudomonas aeruginosa), and 6 other laboratories tested a collection of 55 strains of gram-positive cocci (10 enterococci and 45 staphylococci) against 10-20 antimicrobial agents. The strains were selected on the basis that they harbored challenging and characterized mechanisms of resistance. In comparison with the agar reference method, the automated system gave an overall essential agreement (+/-1 dilution) of 94.5%, 93.5%, and 97% for the gram-negative bacilli, enterococci, and staphylococci, respectively. According to the interpretive standards of the National Committee for Clinical Laboratory Standards, the category agreement ranged from 96 to 96.4% for the three sets of organisms. The accuracy of the automated system, as determined by the kappa test, ranged from 0.80 to 0.88, reflecting an almost perfect agreement with the reference technique. Very major, major, and minor errors obtained with the automated system were 0.3%, 2.9%, and 6.6% for gram-negative bacilli, 3.4%, 0%, and 5% for enterococci, and 1%, 1.6%, and 2.7% for staphylococci, respectively. The high rate of very major errors in enterococci was mostly due to a single strain of multidrug-resistant Enterococcus faecium, which was found susceptible to several antibiotics in a majority of participant laboratories. The use of a heavy inoculum and of a broth test medium by the automated system might account for a better expression of certain resistance mechanisms, including beta-lactamases, as compared to the agar dilution reference method. The interlaboratory reproducibility was acceptable, as shown by the narrow dispersion of MICs and by the results of quality control.