The development of nucleic acid-based technologies has improved the sensitivity, specificity and speed of detection of Mycobacterium tuberculosis in clinical samples. Both commercially available and 'in-house' polymerase chain reaction (PCR) systems are in use, and a significant number of reports compare such systems with more traditional diagnostic tools for tuberculosis. Few studies, however, have focused on the reproducibility of the results when submitting a sample batch to PCR in different laboratories, especially in developing countries. Consequently, PCR results obtained from six laboratories in six different Latin American countries for samples reconstituted with defined amounts of M. tuberculosis cells were evaluated. Each laboratory used specific conditions of sample processing, nucleic acid amplification and amplicon detection. Analysis of results allowed large differences in sensitivity and specificity to be observed. We conclude that in its present setting, in-house PCR cannot be used as a single diagnostic tool for tuberculosis, and that special care needs to be taken upon interpretation of results by inclusion of a proper number of positive and negative controls.