In stress test ECG analysis, the so-called ST/HR hysteresis has recently been suggested to improve coronary artery disease (CAD) diagnosis. This parameter is estimated from the ST versus HR diagram including exercise and recovery phases. Unluckily, ST measurements are adversely affected by noise during the test. In this study we propose a method to automatically estimate the ST/HR hysteresis, incorporating multiple stage noise attenuation. The method is based on averaging and rejection of noisy beats. Evaluation is done on simulated exercise test recordings, constructed from real ECG averaged beats adding actual noise from stress test records. Results on a total of 216 different records, with RMS noise levels ranging from 114 to 979 μV, give a reduction in estimation error in the ST/HR diagram of 77.98% (from 168 to 37 μV) in mean and of 76.38% (from 271 to 63 μV) in standard deviation. This method may be considered as a suitable and robust tool for reliable ST/HR hysteresis estimation.