Abstract : In this work, we study the estimation and the analysis of the cardiac intervals. The main objective of this thesis is therefore to develop new signal processing tools for estimating in ECG, the P-R intervals which are poorly studied in the literature because of the difficulty in extracting the P waves, and the Q-T intervals. Inspired by the method of Charles D. Woody (1967) based on iterative correlation and averaging techniques, we propose firstly an improvement of this method in terms of optimality. The observation model of this method is then enriched by a potentially parasite wave added to noise, and a generalization of the Woody's method is proposed. Taking the example of the T wave of the ECG which overlaps the following P wave when the heart rate is high, various models of the T wave are proposed. With these two proposed methods, the objectives to estimate and analyze cardiac intervals, such as the R-R, Q-T and P-R intervals, are reached. More precisely, the estimation of the P-R intervals on ECG recorded during exercise is feasible applying the generalization of the Woody's method. The application of these methods to real recordings of different types of ECG (rest, exercise, atrial fibrillation,...), has revealed new findings about the evolution of cardiac intervals, as an hysteresis phenomenon in the relation P-R/R-R intervals, or a characterization of subjects according to their training level studying the P-R interval slope in the early recovery phase, for instance. A modeling of the response of the Q-T interval to R-R intervals changes has also been proposed based on the electrical behavior at the cellular level.