The hypothesis that serial assessment of left ventricular function during exercise radionuclide angiography provides improved diagnostic criteria for coronary artery disease (CAD) was examined. Fifty-eight consecutive patients without previous myocardial infarction were prospectively scheduled for cardiac catheterization and multistage radionuclide angiographic exercise studies. Forty-one patients had significant CAD. The traditional criterion--failure to achieve a 5% increment in ejection fraction (EF) during exercise compared with the value at rest--had 85% sensitivity but only 41% specificity for CAD. In 12 patients, EF increased early in exercise by at least 4% and then decreased a mean of 7.5%, often with worsening regional wall motion. This "up-down" EF pattern was applied as a diagnostic test in the overall study group. Analysis of changes in EF from the maximal value achieved to that at the end of exercise resulted in criteria with greater sensitivity (p less than 0.0001) for CAD than analysis of changes from rest, with similar specificity. Regional wall motion abnormalities occurring during the first exercise stage resulted in 94% specificity for CAD (p = 0.05 vs end-stage analysis), although sensitivity was low. Analyzing the maximal EF during exercise results in improved sensitivity, while analyzing the early onset of regional dysfunction results in high specificity for CAD.