A ligase chain reaction (LCR) DNA amplification method for the molecular diagnosis of Mycobacterium tuberculosis (Abbott LCx MTB) was evaluated in comparison with solid and liquid phase culture on 622 selected samples collected in two large Italian hospitals, of which 310 obtained from HIV-1 positive patients and 312 from HIV-negative individuals. The overall prevalence of mycobacteria by culture was 22% (137/622), and the apparent sensitivity and specificity of LCx vs. culture were 87.6% and 98.2%. Of the 26 culture positive/LCx negative samples, 22 were positive for MOTT and 4 for M. tuberculosis. All 9 samples positive by LCx and negative by culture were classified as true positive by clinical criteria. The final values of sensitivity, specificity, positive predictive value and negative predictive value for LCx rose to 96.8%, 100%, 100% and 99.2%, respectively. The adjusted sensitivity of culture methods was 89.5% for solid phase and 92.7% for Bactec. In view of the high sensitivity on both smear-positive (100%) and smear-negative (92.4%) samples and of the high negative predictive value, the LCR-based amplification method appears suitable as a routine screening method for the rapid diagnosis of M. tuberculosis in high-risk patients.