Underreporting, more than overreporting, is a problem in studies of the effects of alcohol consumption using self-reported data. Numerical examples illustrate that in studies of the effect of alcohol, nondifferential misclassification of alcohol consumption due to underreporting may lead to a bias away from the null value. It may also cause a true threshold level for alcohol to appear as a dose-response relationship. It is shown that the effect of misclassification on effect estimates will depend on the true frequency of abstainers in the studied population.