Data sparseness currently limits gene-environment interaction estimation. To improve effect estimates of gene-environment interactions, we give an overview of one approach, hierarchical modeling, and propose a two-stage hierarchical model. The first stage is a logistic model for the joint effects of the genetic and environmental factors. The second stage regresses the joint effects on genotype-specific enzymatic activity of the environmentally derived substrate. The model is illustrated using a case-control study of adenomas of the large bowel, for which NAT2 genotype and dietary data were collected. The first-stage interactions of dietary components and genotype were regressed on initial conversion rates of dietary heterocyclic amines to aryl nitrenium ions. We fit the hierarchical model by penalized likelihood. Compared to effect estimates from maximum-likelihood logistic regression, hierarchical results are more reasonable and precise. These results lend further support to previous observations that hierarchical regression is preferable to ordinary logistic regression when multiple factors and their interactions are being studied. We propose that hierarchical modeling can act as a bridge between molecular epidemiology studies and laboratory data, combining both efficiently.