OBJECTIVES: To conduct a meta-analysis on the diagnostic accuracy of five screening tests for colorectal cancer (CRC): fecal occult blood test (FOBT), double-contrast barium enema (DCBE), flexible sigmoidoscopy (FSIG), conventional colonoscopy (COL) and computed tomography colonoscopy (CTCOL). METHODS: A literature search was carried out in MEDLINE for each test. Articles were reviewed by two independent reviewers. Inclusion criteria were: 1) RCTs or observational studies of CRC screening; 2) patients with low/average risk of CRC; 3) complete data to calculate sensitivity and specificity. Exclusion criteria were: 1) non-peer reviewed articles; 2) articles whose primary aim was not to assess CRC screening; 3) articles not in English/French; 4) articles published prior to 1975; and 5) high risk screening populations. Weighted linear regression was used to identify significant covariates. Sensitivity and specificity were pooled for relevant subgroups. RESULTS: The initial literature search found 399 articles for FOBT, 253 for DCBE, 394 for FSIG, 434 for COL, and 345 for CTCOL. Of these, 12, 8, 10, 8, and 13 articles respectively, were included in the final analysis. With the exception of colonoscopy the remaining tests showed evidence of heterogeneity and threshold effect. Significant covariates included study design and type of FOBT. Pooled sensitivity and specificity (95% CI) for randomized FOBT and FSIG trials were 0.738 (0.705,0.768), 0.960 (0.959,0.961), 0.822 (0.770,0.864), and 0.997 (0.994,0.998) respectively. For the remaining tests, the pooled sensitivity and specificity (95% CI) were 0.767 (0.728,0.802), 0.975 (0.970,0.979) for all DCBE studies, 0.867 (0.828,0.898), 0.995 (0.991,0.998) for all COL studies and 0.879 (0.840,0.910), 0.964 (0.956,0.971) for all CTCOL studies. CONCLUSIONS: When heterogeneity is present within test groups, results from pooled sensitivity and specificity can be misleading. A planned future step is to estimate diagnostic odds ratios and build summary ROC curves which are more reliable estimates of test accuracy for evidence synthesis.