Although exercise radionuclide ventriculography was initially reported to be a highly specific test for coronary-artery disease, later studies reported a high false-positive rate. To verify this turnabout, we analyzed the responses in 77 angiographically normal patients; 32 were studied from 1978 to 1979 (the early period), and 45 from 1980 to 1982 (the recent period). Most patients studied in the early period had normal responses (94 per cent for ejection fraction and 84 per cent for wall motion). In contrast, normal responses were less frequent in patients studied in the recent period (49 per cent for ejection fraction and 36 per cent for wall motion, P less than 0.001). The probability of coronary disease before testing was higher in these patients (38 vs. 7 per cent, P less than 0.001). More patients studied in the recent period underwent radionuclide ventriculography before angiography (78 vs. 22 per cent, P less than 0.001), and more of these prior studies had abnormal results than those performed after angiography (55 vs. 6 per cent, P less than 0.0001). Thus, two factors are responsible for the temporal decline in specificity: a change in the population being tested (pretest referral bias) and a preferential selection of patients with a positive test response for coronary angiography (post-test referral bias).