The effect of misclassification of Wolfe's mammographic classifications was investigated using data from two radiologists. If there are only two risk groups (high and low), the apparent relative risk expected from typical surveys would be around 2 or 3, even if the true relative risk is very high. If each of Wolfe's four classes has its own risk, the difference between N1 and P1, on the one hand, and P2 and DY on the other, would be almost lost due to misclassification. Published surveys in which the observed relative risks for the high-risk group (P2 and DY) are low are not inconsistent with Wolfe's original finding of a high relative risk: misclassification can greatly distort estimates of relative risks. If misclassification can be managed by restricting uncertain mammograms to one class (or by removing them from the study altogether), the bias can be greatly reduced.