Lung cancer association studies have yielded suggestive but not definitive results for a few genes: CYP1A1 (in Japanese), GSTM1 and CYP2D6. We focus on variability in studies of lung cancer and the CYP2D6 gene (debrisoquine metabolic phenotype) as an instructive case and we propose some sources for this heterogeneity. Beyond the general sources of bias in all field studies, three specific concerns are relevant. First, evidence that CYP2D6 is expressed in the brain. The metabolic phenotypes have distinct psychological profiles and therefore there is the potential to distort studies through selection bias. Second, among the lung cancer histologic types, adenocarcinoma likely does not share increased genetic susceptibility due to CYP2D6. Third, the degree of smoking is likely to be related to genetic susceptibility. Effect modification by smoking should be sought for any putative genetic marker for lung cancer. Progress in understanding genetic susceptibility is likely to depend on future well-designed studies that adjust for these and other sources of bias. We are currently reanalyzing the original data from the published studies in order to further explore these issues.