Several indexes have been reported to improve the accuracy of exercise test electrocardiogram (ECG) analysis in the diagnosis of coronary artery disease (CAD), compared with the classical ST depression criterion. Some of them combine repolarisation measurements with heart rate (HR) information (such as the so-called ST/HR hysteresis); others are obtained from the depolarisation period (such as the Athens QRS score); finally, there are heart rate variability (HRV) indexes that account for the nervous system activity. The aim of this study was to identify the best exercise ECG indexes for CAD diagnosis. First, a method to automatically estimate repolarisation and depolarisation indexes in the presence of noise during a stress test was developed. The method is divided into three stages: first, a preprocessing step, where QRS detection, filtering and baseline beat rejection are applied to the raw ECG, prior to a weighted averaging; secondly, a post-processing step in which potentially noisy averaged beats are identified and discarded based on their noise variance; finally, the measurement step, in which ECG indexes are computed from the averaged beats. Then, a multivariate discriminant analysis was applied to classify patients referred for the exercise test into two groups: ischaemic (positive coronary angiography) and low-risk (Framingham risk index < 5%). HR-corrected repolarisation indexes improved the sensitivity (SE) and specificity (SP) of the classical exercise test (SE = 90%, SP = 79% against SE = 65%, SP = 66%). Depolarisation indexes also achieved an improvement over ST depression measurements (SE = 78%, SP = 81%). HRV indexes obtained the best classification results in our study population (SE = 94%, SP = 92%) by means of the very high-frequency power (VHF) (0.4-1 Hz) at stress peak.