OBJECTIVE--To examine how exercise testing on background medical treatment affects the ability of the test to predict prognostically important patterns of coronary anatomy in patients with a high clinical probability of coronary artery disease but who are well controlled on medication. DESIGN--Prospective study. SETTING--Regional cardiothoracic centre and referring district general hospital. PATIENTS--84 patients with a history of typical angina or definite myocardial infarction and mild symptoms who had been placed on the waiting list for prognostic angiography. INTERVENTION--Maximal exercise electrocardiography and radionuclide ventriculography performed off and on medication, followed by angiography within three months. MAIN OUTCOME MEASURE--Prognostically important coronary artery disease for which early surgery might be recommended purely on prognostic grounds, irrespective of symptoms. RESULTS--Coronary artery disease was present in 71/84 (85%) patients; in 28/84 (33%) patients this was prognostically important. When the result was strongly positive, the predictive accuracy for prognostically important disease was 0.46 off and 0.62 on medication for the exercise electrocardiogram and 0.71 off and 0.82 on medication for exercise radionuclide ventriculography. The likelihood ratio was 1.00 off and 1.36 on medication for exercise electrocardiography and 2.54 off and 10.5 on medication for exercise radionuclide ventriculography. In stepwise logistic regression, the test identified as the strongest predictor of prognostically important disease was exercise radionuclide ventriculography on medication for which the improvement chi 2 was 28 (p < 0.0001). With the regression model, the probability of important disease is 92% if exercise radionuclide ventriculography on medication is at least strongly positive, compared with 16% if the result is normal or just positive. CONCLUSION--In patients likely to have coronary disease, exercise testing should be performed without interruption of medication to optimise its ability to identify those with prognostically important disease, and to help to avoid unnecessary or premature angiography in those who are well controlled on medical treatment.