Since 1981, numerous epidemiological studies have investigated the relationship between passive smoking and lung cancer in nonsmokers. The overall evidence, predominantly relating to women, indicates a weak association with the husband's smoking and many reviewers have concluded that this demonstrates a causal effect of exposure to environmental tobacco smoke (ETS). Interpreting weak associations is notoriously difficult, however, and this paper reviews problems specific to the ETS-lung cancer relationship. After describing how to select relevant studies and appropriate data, the methods for combining evidence together ('meta-analysis') are discussed, and the need to investigate sources of heterogeneity is emphasized. Separate consideration is given to various forms of bias that may affect overall relative risk estimates, including misclassification of active smoking status, confounding, systematic case-control differences, recall bias, diagnostic bias and publication bias. Sections on dose-response, multiple ETS exposure sources and other issues follow. The problems are illustrated from the available literature. It is shown there is no significant association of lung cancer with workplace, childhood or social ETS exposure or with smoking by the wife. Though statistically significant, the association with husband's smoking is weak and heterogeneous and varies widely according to various study characteristics. The association is markedly weakened by the adjustment for smoking misclassification bias and is likely to be affected by confounding and other sources of bias. While the precise extent of all the biases remains unclear, it seems impossible to conclude with any certainty that ETS causes lung cancer.