The exercise electrocardiogram (ECG) is frequently used as a screening tool for coronary artery disease (CAD) in epidemiological studies and routine health evaluation. Estimates of the validity of this diagnostic test have usually been obtained by correlating the exercise ECG with the results of coronary angiography. However, coronary angiography is usually performed only on selected individuals with symptomatic CAD. If symptomatic subjects are more likely to respond positively to the exercise ECG than asymptomatic individuals with comparable severity of disease, then biased estimates of the test validity for an unselected population may be obtained. In this paper the validation group bias is shown to be a function both of the proportion of the population with symptomatic CAD and of the relative validity of the test for the symptomatic and asymptomatic subjects. The extent of this bias is examined.