Validation studies are increasingly employed to estimate and correct for exposure misclassification in epidemiologic studies. Often, however, the reference procedure used for validation is less than perfect. This phenomenon may lead to biased estimates of misclassification rates and hence of "corrected" measures of associations. In particular, it has been demonstrated that basic matrix correction procedures to correct for presumably nondifferential exposure misclassification may yield strongly misleading results in the case of an imperfect gold standard. The problem arises because apparent misclassification (misclassification with respect to an alloyed gold standard) is often differential even if true misclassification (misclassification with respect to true exposure status) is nondifferential. Therefore, correction procedures for differential rather than nondifferential misclassification should be employed in the presence of an alloyed gold standard.