Recognition of patients with high risk for ventricular tachycardia (VT) or sudden cardiac death is of high clinical importance. We have investigated the efficiency of maximum entropy spectral estimation (MES) to detect such risk patients on the basis of highly amplified surface ECG. In comparison with the traditionally applied periodogram (fast Fourier transform), the MES produces sharper and more pronounced peaks in the power density spectrum (PS). The main problem is the influence of residual noise (after averaging), which often leads to additional components in the PS. To completely avoid this negative noise influence we developed a new algorithm, called the variance subtraction method. In a first clinical investigation 86 per cent of patients with myocardial infarction and ventricular tachycardia have shown frequency components above 80 Hz in the PS compared with healthy persons where no frequency components above this 80 Hz level could be detected.