The evidence for the adverse health effects of today's "normal" air pollution relies heavily on epidemiologic studies. Given the major impact of these effects on environmental, economic and social policy, the causality of epidemiologic findings becomes a central issue. The scientific process of establishing causality based on the air pollution epidemiology literature is reviewed. It consists of two steps: first, the quality and validity of each epidemiologic study must be rigorously assessed (design, selection, confounding); second, the overall evidence across all valid studies must be appraised. Of major importance are the consistency of results between replicate studies and the coherence of effects across a variety of different pathophysiologically or logically interrelated epidemiologic health outcomes, such as respiratory symptoms, exacerbations of respiratory diseases, use of health care, sick absenteeism, or, among the most sensitive subjects, mortality. Demonstration of a dose-response association is crucial and has been shown in many studies. Valid studies may use varied indicators of the air pollution mixtures, such as particulate matter, NO2, SO2, or others. However, epidemiology is incapable of investigating the biologic pathophysiologic mechanisms, which require experimental studies. Such studies may increase the biologic plausibility of epidemiologic findings. Given the common sources of many of the air pollution indicators used in epidemiology, a knowledge of pathophysiologic mechanisms is not a prerequisite for public health action aimed at improving air quality in cities.