A large epidemiologic study on residential radon exposure and lung cancer has been conducted in Sweden. An attempt is now made to quantify the impact of random error in the exposure assessment on the risk estimate for lung cancer in this study. The study included 1,360 lung cancer cases, diagnosed from 1980 to 1984, and 2,847 population controls. Radon measurements were performed in 8,992 dwellings occupied by the study subjects some time since 1947. Questionnaires provided information on smoking and other risk factors. Imprecision in the retrospective exposure assessment for radon was estimated from a Monte Carlo technique modeling Swedish conditions. Adjusted risk estimates were obtained from regression analyses based on expected values for true time-weighted average residential radon concentration (TWA), conditional on observed TWA. Without adjustment for random error in the TWA estimates, the linear excess relative risk coefficient was 0.10 per 100 Bq m(-3), but an excess relative risk of about 0.15 to 0.20 per 100 Bq m(-3) was suggested following adjustment. The potentially significant consequences of errors in the retrospective radon exposure assessment should be taken into consideration in the risk estimation as well as in comparisons of results of different studies and in future pooled analyses.