The risk estimates obtained in studies of genetic polymorphisms and lung cancer differ markedly between studies, which might be due to chance or differences in study design, in particular the stratification/match of comparison group. The effect of different strategies for stratification and adjustment for smoking on the estimated effect of polymorphisms on lung cancer risk was explored in the case-cohort design. We used an empirical and a statistical simulation approach. The stratification strategies were: no smoking stratification, stratification for smoking status and stratification for smoking duration. The study base was a prospective follow-up study with 57,053 participants. In the simulation approach the glutathione S-transferase T1 null polymorphism, as a model of any polymorphism, was added to simulated data in two different ways, assuming either absence or presence of association with smoking. In the empirical approach the risk estimates of the investigated polymorphisms differed between the three different stratification strategies. Simulated data with neither stratification nor adjustment for smoking resulted in low biases and narrow confidence intervals (CI) in the absence of a genotype-smoking association and markedly higher biases in the presence of a genotype-smoking association. In study designs stratified by smoking, low biases and narrow CI spans were found, regardless of a genotype-smoking association. Stratification for smoking seems to be advantageous in case-cohort studies of genetic polymorphisms and lung cancer.