The aim of the study was to review published articles on the diagnostic accuracy of C-reactive protein (CRP) tests with cerebrospinal fluid and serum in diagnosing bacterial meningitis. The literature from 1980 and onwards was searched using the electronic databases of MEDLINE, and we used summary receiver operating characteristic curve analyses (SROCs) to describe central tendencies and examine possible sources of inter-study variability in the results. We included data from 35 studies of both children and adults: 21 in which CRP had been measured in cerebrospinal fluid, 10 in which CRP had been measured in serum, and 4 in which it had been measured in both cerebrospinal fluid and serum. The odds ratio for bacterial meningitis versus aseptic meningitis for a positive CRP test with cerebrospinal fluid was estimated at 241 (95% confidence interval [CI]: 59-980), and the central tendencies for the true-positive fraction (sensitivity) and the false-positive fraction (1-specificity) were estimated at 0.94 and 0.06, respectively. The corresponding figures for a CRP test with serum were 150 (95% CI: 44-509), 0.92 and 0.08, respectively. Regression analyses including variables coding for study design features, inclusion of neonatal patients, geographical region, or use of a quantitative biochemical method did not indicate statistically significant contributions to inter-study variances in the log odds ratios. For values of the true- and the false-positive fractions of 0.92-0.94 and 0.06-0.08, respectively, the post-test probability of not having bacterial meningitis given a negative test is very high (> or = 97%), in the range of a pre-test probability (prevalence of bacterial meningitis) from 10 to 30%, whereas the post-test probability of bacterial meningitis given a positive test is considerably lower. Hence, only a negative test is highly informative in a typical clinical setting. This, as well as the absence of analyses to show if CRP tests contribute independent diagnostic information, relatively to the information held in the traditionally used clinical and biochemical variables, makes it difficult to conclude on the clinical usefulness of CRP tests in the management of patients suspected of having bacterial meningitis.