We describe two statistical designs that can provide efficient estimates of the health effects of exposure to air pollutants in epidemiologic studies. We also evaluate the effects of measurement error in exposure assessment on the accuracy of estimated health effects. The bidirectional case-crossover design is a variant of a method proposed by Maclure (1991). Our version of the method takes advantage of the fact that in epidemiologic studies involving environmental exposure, accurate information about past exposure is more readily available, and that levels of exposure are generally unaffected by the response of the subject. It differs from other case-crossover methods in that control information is assessed both before and after failure, thus avoiding confounding due to time trends in exposure. The multilevel analytic design provides a method of combining estimates of health effects made on the individual level with those made at the group level. It has great potential value in situations where variations in exposure within groups may not be great enough to provide adequate power to detect health effects, as is often the case in air pollution studies where exposure levels are similar within a geographic community. Measurement errors in exposure assessment can have substantial impact on the accuracy of estimated health effects. When the microenvironmental approach is used to estimate exposure, a standard error of 30% in estimating indoor/outdoor ratios can increase the standard error of a relative risk estimate by 50%, and introduce bias as well. Similar results hold when exposure is estimated with personal samplers. When the microenvironmental approach is used, errors in estimating indoor/outdoor ratios have more influence on the accuracy of risk estimation than do errors in estimating the time spent in microenvironments.