Although the signal-averaging technique is over one hundred years old, its first use was demonstrated on the human heart in 1963 and ten years later recording of His bundle potentials from the body surface became feasible. During the last decade interest has been focused on ventricular late potentials (LP) which represent zones of delayed myocardial activation occurring in hearts after myocardial infarction as well as in diffusely damaged myocardium. Presentation of LP in the 'time domain analysis' is highly standardized and widely accepted. Development of spectral analysis techniques is still advancing and promises not only a more sophisticated interpretation of LP but also detection of electrical disturbances during the whole QRS complex. In clinical cardiology LP are found in 10% to 50% of all cardiac diseases, mainly in coronary heart disease and dilated cardiomyopathy. About 80% of patients with sustained ventricular tachyarrhythmia demonstrate LP. Their value as predictive of arrhythmic events and sudden cardiac death is relatively low (10-30%), but very high in predicting a good outcome, showing 95% event free survival if LP are negative. Predictive accuracy can be enhanced by a combination of independent indices, particularly heart rate variability. Other clinical applications are prediction of patency of the infarct-related artery after thrombolysis or after angioplasty in acute myocardial infarction and prediction of success after guided anti-tachycardiac surgery. No consistent changes can be observed following elective revascularization by angioplasty or bypass surgery. Ischaemia, exercise and manipulation of autonomic tone apparently do not consistently influence LP if ensemble signal-averaging is applied, but may cause transient alterations detectable by beat-to-beat analysis. Potential future directions are (1) evaluation of spectral analysis techniques, (2) improvement of the predictive value by combining time and frequency domain analysis, (3) increased application of beat-to-beat analysis in different pathophysiological conditions, (4) enhanced diagnosis of high-risk patients by combining other independent arrhythmic indices with LP, (5) specific selection of patients for invasive electrophysiological testing and (6) the study of LP shifting during long-term therapy of heart hypertrophy, failure, inflammation and ischaemia. Dynamic or functional LP analysis during exercise testing, electrophysiological investigation or coronary angioplasty may be a further application.