Signal processing techniques were developed to analyze hum and noise, the two main factors influencing the quality of exercise electrocardiograms. Hum was accurately evaluated by Fourier analysis methods, and the noise characteristics by appropriate sampling procedures and autocorrelation analysis. In a sample study of 41 typical exercise ECG records, the average hum over a 3-minute record was 34 μv peak to peak with a minimum of 4 μv peak to peak and a maximum of 340 μv peak to peak. Noise was found to closely approximate Markoffian “random” noise plus periodic components closely correlated to both the periodicity of the exercise test and of respiration. The total noise varied from a root-mean-square value of 60 μv to a maximum of 915 μv. The periodic noise power averaged 8 per cent of the total noise power, with a minimum of 1 per cent to a maximum of 30 per cent of the total power. Pilot studies were made to determine differences in the noise content due to slight variations in the anatomic position of the electrodes, and due to different stress tests. Significant and consistent variations were seen in noise content of electrodes whose position varied slightly. Similarly, it was determined that the average standard deviation of the noise in the Master two-step test was about 50 per cent higher than that from a bicycle ergometer test. An improvement in the average signalto-noise ratio was found by means of spatial averaging, i.e., linear weighting of the signals from electrodes located very close together anatomically. This “realtime” averaging also reduced the statistical chance that a single bad electrode would ruin an entire record. Also, of considerable interest are the encouraging results when multiple redundancy of leads was employed. Here, instead of continuously averaging all 3 signals from the closely spaced electrodes, all signals were continuously monitored and the best combination of 2 leads was selected. This selected combination gave noise figures that were better than the best single lead in all but 3 out of 19 test runs.