When misclassification of exposure and disease is nondifferential but not independent of one another, bias away from the null can result. For dichotomous variables, misclassification is nonindependent when the probability of misclassification of one variable is dependent on the correctness of classification of the other variable. One plausible form of nonindependent misclassification may result from variation in the threshold for reporting exposure and outcome by study subjects. The odds ratio after dependent misclassification can be expressed as a function of the true odds ratio, the prevalences of exposure and outcome, and the probabilities of misclassification. When prevalences of exposure and outcome are low, bias may be considerable even at low probabilities of misclassification. The nonindependent misclassification described in this article will result in a positive bias in the odds ratio and is therefore of prime concern when questioning the validity of an observed effect. The core of the problem lies in the study design and can be solved by eliminating the common link that makes nonindependent errors possible.