A test is often interpreted as "normal" or "abnormal" by a single criterion, regardless of the intent of testing. The discriminate accuracy of this convention was critically analyzed using information content (I), likelihood ratio and the area under the receiver-operating characteristic curve. Three ejection fraction variables were assessed--ejection fraction at rest, exercise ejection fraction and the change in ejection fraction from rest to exercise--each relative to three intentional goals: diagnosis of coronary artery disease in 929 patients without previous myocardial infarction, prediction of multivessel disease in these same 929 patients and prediction of multivessel disease in 507 patients with previous myocardial infarction. The information content of exercise ejection fraction (IEX) was higher than for ejection fraction at rest (IR) or for the change from rest to exercise (IEX-R), and was relatively constant regardless of the goal of testing. In contrast, neither IR nor IEX-R was constant. IR was lowest for diagnosis of coronary artery disease, whereas IEX-R was highest for this same goal. These empiric observations are consistent with the quantitative relation predicted by information theory: IEX = IR + IEX-R. Thus, ejection fraction at rest has little discriminate value relative to the diagnosis of coronary artery disease, but does have value in evaluating the extent of disease in patients after myocardial infarction. Exercise ejection fraction and change in ejection fraction are nearly equally useful for purposes of diagnosis, whereas the former is most useful for functional evaluation in postinfarction patients.(ABSTRACT TRUNCATED AT 250 WORDS)