To investigate reasons for the wide variation in formal studies of sensitivity and specificity indexes for the diagnostic efficacy of the graded exercise test for angiographically defined coronary disease, data were collected on 205 consecutive exercise tests at two hospital-based exercise laboratories. For calculations of sensitivity and specificity, stress test data are usually analyzed with many exclusions for ineligibility, with equivocal results omitted, and only in patients undergoing angiography. Consequently, only 3% of patients who received the tests in this survey would have been included in a typical formal study of diagnostic efficacy. In the same way that the visible tip of an iceberg misrepresents its extent and depth, the patients assembled in studies of diagnostic tests may be a highly selected group that misrepresents the intended population.