The diagnostic performance of automatic analysis of the exercise electrocardiogram in detecting ischaemic heart disease was studied in 147 patients with angiographically documented coronary disease. The results were compared with the results of visual analysis of the same recordings. Using a bicycle ergometer we tried to reach at least 90 per cent of the predicted maximal heart rate of the patient. Two bipolar thoracic leads (CM5, CC5) were used. In the visual analysis the criterion of the so-called ischaemic ST segment was applied. For the automatic analysis the population was divided into a learning group (N=87) and a testing group (N=60). In the learning group first critical values were computed for different ST measurements that provided optimal separation between patients with (CAG POS.) and without (CAG. NEG.) significant coronary stenoses as revealed by coronary arteriography. These critical values were kept unchanged when applied to the testing group. With respect to the visual method an increase of the sensitivity by 0-45 and 0-36 was obtained by the automatic analysis in the learning and testing group, respectively. The best separation between CAG. POS. and CAG. NEG. group was reached using a criterion consisting of a linear combination of the slope of the initial part of the ST segment and the ST depression; the sensitivity being 0-70 and 0-60, respectively, in the learning and testing group. Using a criterion based on the area between the baseline and the ST segment (the SX integral) these values were 0-42 and 0-49, respectively. All specificities were kept to at least 0-90.