We used the case-crossover design to identify any increase in mortality in Seoul, Korea, when there were higher levels of ambient air pollution on case-days than would be expected solely as a result of chance. This empirical study showed that either unidirectional retrospective (selecting only control days prior to death) or prospective (selecting only control days after death) control sampling could cause risk estimates to be confounded by seasonal waves as well as time trends in air pollution levels. In bidirectional control sampling in which exposures at death were compared with exposures both before and after death, the estimated mortality was resistant to confounding by time patterns of air pollution. Using a bidirectional control sampling approach, the results from a conditional logistic regression model controlling for weather conditions showed that the nonaccidental mortality associated with a 50-ppb increment over a 3-day moving average of SO(2) concentrations, including the concurrent day and preceding 2 days, was 1.023 [95% confidence interval (CI), 1.016-1.084]. The relative risk of death was 1.023 (CI, 0.999-1.048) per 50 ppb for 1-hr maximum O(3) and 1.010 (CI, 0.988-1.032) per 100 microg/m(3 )or total suspended particulates. In conclusion, the findings of this study were 2-fold: given the consistency of the observed association between SO(2) and daily mortality across different analysis methods, the association reported here indicates that air pollution is a probable contributor to premature death; and bidirectional control sampling is needed in a case-crossover design applied to air pollution epidemiologic studies to control confounding by seasonal patterns of air pollution as well as time trends.