To determine why exercise testing remains controversial as a diagnostic test for coronary artery disease, a methodologic review was undertaken of 33 studies comprising 7,501 patients who had undergone both exercise tests and coronary angiography. Of seven methodologic standards for research design, only one received general compliance: the requirement for an adequate variety of anatomic lesions. Less than half of the studies complied with any of the remaining six standards: adequate identification of the groups selected for study; adequate analysis for relevant chest pain syndromes; avoidance of a limited challenge group; and avoidance of work-up bias, diagnostic review bias and test review bias. Only one study met as many as five standards. These methodologic problems may explain the wide range of sensitivity (35 to 88 percent) and specificity (41 to 100 percent) found for exercise testing, because the variations could not be attributed to the usual explanations: definition of anatomic abnormality, stress test technique or definition of an abnormal test. Determining the true value of exercise testing requires methodologic improvements in patient selection, data collection and data analysis.