Nucleic acid amplification to detect Mycobacterium tuberculosis in clinical specimens is increasingly used as a laboratory tool for the diagnosis of tuberculosis. However, the specificity and sensitivity of these tests may be questioned, and no standardized reagents for quality control assessment are available. To estimate the performance of amplification tests for routine diagnosis, we initiated an interlaboratory study involving 30 laboratories in 18 countries. We prepared blinded panels of 20 sputum samples containing no, 100, or 1,000 mycobacterial cells. Each laboratory was asked to detect M. tuberculosis by their routine method of nucleic acid amplification. Only five laboratories correctly identified the presence or absence of mycobacterial DNA in all 20 samples. Seven laboratories detected mycobacterial DNA in all positive samples, and 13 laboratories correctly reported the absence of DNA in the negative samples. Lack of specificity was more of a problem than lack of sensitivity. Reliability was not found to be associated with the use of any particular method. Reliable detection of M. tuberculosis in clinical samples by nucleic acid amplification techniques is possible, but many laboratories do not use adequate quality controls. This study underlines the need for good laboratory practice and reference reagents to monitor the performance of the whole assay, including pretreatment of clinical samples.