Observational studies and randomized trials provide relevant and complementary information to a totality of evidence on which to base rational clinical decision making for patients and overall health policy for the general population. Observational studies are particularly useful for detecting moderate to large effects. The 15- to 20-fold greater risk of lung cancer among long term cigarette smokers was established by case-control and prospective cohort studies. The approximate 80% increased risk of coronary heart disease associated with current smoking also has been reliably demonstrated in observational studies. However, as the relative risk gets smaller, there is increasing concern that unmeasured or unknown confounding variables may account for all or part of any observed association. For these reasons, reliable inferences about interventions likely to confer small to moderate benefits will emerge only from randomized trials of sufficient sample size and duration of treatment and follow-up. Dietary variables have been postulated to account for as much as 35% of all human cancers. However, the hypothesized benefit of any specific dietary constituent, such as the antioxidant beta-carotene, is likely to be modest in size, on the order of a 20-30% reduction in risk. Therefore, although a large number of observational studies have demonstrated that individuals with higher dietary intakes or blood levels of beta-carotene have lower risks of cancer, only randomized trials can address this hypothesis definitively. Such trials, however, must be of sufficient duration to allow for the development of an anticancer effect. This may mean a decade or more based on the analogy with smoking cessation and decreased risks of lung cancer. Several ongoing large-scale trials are testing beta-carotene and other promising cancer chemoprevention agents, and their results will provide clear evidence on the balance of benefits and risks of these interventions.