Although variation in diet may account for approximately one third of the variation in cancer incidence worldwide, epidemiologic studies have proven to be a blunt instrument for identifying causal relationships between intakes of specific food constituents and cancer risk at specific sites. Diets rich in fruits and vegetables seem to be protective, but the adverse effects of beta carotene supplementation trials on lung cancer incidence in smokers caution against the attribution of benefit to single substances. Important diet–gene interactions may exist, as illustrated by differential responses to variation in folate status in those with methylenetetrahydrofolate reductase polymorphisms. Targeting initial intervention studies in those with explicit genetic predisposition to cancer may have both greater cost-effectiveness and fewer ethical difficulties than do similar studies in the general public.