Abstract : A new filtering technique for high-resolution ECG signals processing is proposed. The filter is based on approximation of the electrocardiogram segments by polynoms of low degree with possibility of including smoothing constraints at the frontier points. The segments localization, the polynomial degree and the continuity constraints are discussed in details. Finally it is shown that the proposed filter, without continuity constraints is FIR, linear and shift invariant. The filter may find application in analysis of the heart micropotentials, especially late potentials (LPs). In this case the idea is to match the filter to the high amplitude segments of the ECG in order to limit their influence in the part of the signal where micropotentials are expected. The preliminary results obtained for post-infarction patients show that the filtration allows for clear detection of abnormal electrical activity in the terminal part of the QRS complex.