Abstract In this study of a canine heart model of localized reversible ischemia, a computer-based signal-processing method is developed to detect and localize the epicardial projections of ischemic myocardial electrocardiograms (ECGs) during the cardiac activation, rather than the repolarization, phase. This is done by transforming ECG signals from an epicardial sensor array into the multichannel spectral domain and identifying three decision variables: (1) the frequency in hertz of the spectral peak (f 0), its frequency band width 50% below the peak value (w 0), and the maximum eigenvalue difference of the ECG signal's autocorrelation matrix (e 0). With use of the histograms of the f 0, w 0, and e 0 parameters of 3256 ECGs from normal and 957 from ischemic areas of myocardium obtained from 12 dogs, it was possible to predict ischemia in a new test group of nine animals from a Neyman-Pearson (NP) test in which the threshold probabilities of detecting ischemia for each decision variable are compared with those of detecting normality. Quantification of each sensor area by the NP tests revealed that, compared with the control, ECG spectra with decreased F 0 and w 0 and increased e 0 relative to their respective thresholds had increased myocardial laclate (p < 0.01), decreased adenosine triphosphate (ATP) (p < 0.05), and reduced creatine phosphate (p < 0.01). Prediction of f 0 (p < 0.0006) as a continuous variable could be obtained from the regression of the myocardial levels of ATP plus creatine phosphate, which demonstrated that this decision variable appears to directly reflect myocardial energetics. It appears that an advanced signal-processing method for ECG array data can be used to detect, localize, and quantify reversible myocardial ischemia.