We systematically reviewed the current understanding of human population immunity against SARS-CoV in different groups, settings and geography. Our meta-analysis, which included all identified studies except those on wild animal handlers, yielded an overall seroprevalence of 0·10% [95% confidence interval (CI) 0·02–0·18]. Health-care workers and others who had close contact with SARS patients had a slightly higher degree of seroconversion (0·23%, 95% CI 0·02–0·45) compared to healthy blood donors, others from the general community or non-SARS patients recruited from the health-care setting (0·16%, 95% CI 0–0·37). When analysed by the two broad classes of testing procedures, it is clear that serial confirmatory test protocols resulted in a much lower estimate (0·050%, 95% CI 0–0·15) than single test protocols (0·20%, 95% CI 0·06–0·34). Potential epidemiological and laboratory pitfalls are also discussed as they may give rise to false or inconsistent results in measuring the seroprevalence of IgG antibodies to SARS-CoV.