To evaluate the effectiveness of available rapid diagnostic tests to identify tuberculosis (TB) infection. Electronic databases were searched from 1975 to August 2003 for tests for active TB and to March 2004 for tests for latent tuberculosis infection (LTBI). Studies were selected and evaluated that (1) tested for LTBI, (2) compared tuberculin skin test (TST) and interferon-gamma assays based on ESAT-6 and CFP-10 antigens and (3) provided information on TB exposure or bacille Calmette-Guerin (BCG) vaccination or HIV status. For each test comparison, the sensitivity, specificity and 95% confidence intervals (CIs) were calculated. Sources of heterogeneity were investigated by adding covariates to the standard regression model. The authors examined whether interferon-gamma assays were more strongly associated with high versus low TB exposure than TST. Odds ratios (ORs) were calculated for the association between test results and exposures from each study along with their 95% CIs. Within each study, the OR value for one test was divided by that for another to produce a ratio of OR (ROR). A total of 212 studies were included, providing 368 data sets. A further 19 studies assessing fully automated liquid culture were included. Overall, nucleic acid amplification test (NAAT) accuracy was far superior when applied to respiratory samples as opposed to other body fluids. The better quality in-house studies, were, for pulmonary TB, much better at ruling out TB than the commercial tests (higher sensitivity), but were less good at ruling it in (lower specificity), but it is not possible to recommend any one over another owing to a lack of direct test comparisons. The specificity of NAAT tests was high when applied to body fluids, for example for TB meningitis and pleural TB, but sensitivity was poor, indicating that these tests cannot be used reliably to rule out TB. High specificity estimates suggest that NAAT tests should be the first-line test for ruling in TB meningitis, but that they need to be combined with the result of other tests in order to rule out disease. Evidence for NAAT tests in other forms of TB and for phage-based tests is significantly less prolific than for those above and further research is needed to establish accuracy. There is no evidence to support the use of adenosine deaminase (ADA) tests for diagnosis of pulmonary TB; however, there is considerable evidence to support their use for diagnosis of pleural TB and to a slightly lesser extent for TB meningitis. Anti-TB antibody test performance was universally poor, regardless of type of TB. Fully automated liquid culture methods were superior to culture on solid media, in terms of their speed and their precision. In total, 13 studies were included. Assays based on RD1 specific antigens, ESAT-6 or CFP-10, correlate better with intensity of exposure, and therefore are more likely than TST/purified protein derivative (PPD)-based assays to detect LTBI accurately. An additional advantage is that they are more likely to be independent of BCG vaccination status and HIV status. The NAAT tests provide a reliable way of increasing the specificity of diagnosis (ruling in disease) but sensitivity is too poor to rule out disease, especially in smear-negative (paucibacillary) disease where clinical diagnosis is equivocal and where the clinical need is greatest. For extra-pulmonary TB, clinical judgement has both poor sensitivity and specificity. For pleural TB and TB meningitis, adenosine deaminase tests have high sensitivity but limited specificity. NAATs have high specificity and could be used alongside ADA (or interferon-gamma) to increase sensitivity for ruling out disease and NAAT for high specificity to rule it in. All studies from low-prevalence countries strongly suggest that the RD1 antigen-based assays are more accurate than TST- and PPD-based assays for diagnosis of LTBI. If their superior diagnostic capability is found to hold up in routine clinical practice, they could confer several advantages on TB control programmes. Further research for active TB needs to establish diagnostic accuracy in a wide spectrum of patients, against an appropriate reference test, and avoiding the major sources of bias. For LTBI, research needs to address different epidemiological and clinical settings, to evaluate the performance of the main existing commercial assays in head-to-head comparison in both developed and developing countries, and to assess the role of adding more TB-specific antigens to try to improve diagnostic sensitivity.