For over a century, electrocardiology has been observing human cardiac activity through recordings of electrocardiograms (ECG). The potential differences derived from the nine electrodes of the standard 12-lead ECG, placed at their designated positions, are the expression of electric dynamics of which the heart is the source. According to well-defined protocols and established criteria of diagnosis, the signals of the electrocardiogram are used as indicators of cardiac pathology. However, of the four chambers of the human heart, each of which has a specific function, most attention in cardiology has been traditionally placed on the ventricles. This has meant that the conventional ECG system is focused on the observation of ventricular activity, and might not be optimal in studying the activity of the atria. The increasing prevalence of atrial fibrillation in the general population, with its inherent severe complications as well as the known social and economic impacts of the disease, has elicited studies investigating body surface potentials of atrial arrhythmias, invariably pivoted on the standard ECG. The aim of this thesis is to investigate the conception and validation of a lead system targeted at the analysis of atrial fibrillation. This new lead system should be dedicated and optimized to capturing a maximal amount of information about the atrial electric activity taking place during fibrillation, but at the same time be well anchored to the standard ECG configuration, in view of its application in clinical practice. This constraint has led to the use of the same number of electrodes, nine, while leaving at least half of these, five, in their initial positions. In the first part of this thesis, observations of body surface potential maps during normal atrial activity are discussed. The objective was to study the involvement of atrial repolarization in body surface potentials. While studying ECG signals recorded with 64-lead systems from 73 patients, special attention was devoted to the processing of low-amplitude signals. The local potential extremes were found at positions not sampled by the standard leads. Moreover, the PQ segment was found to be not electrically silent, the time course of the potential distribution being very similar to that during the P wave but for a reversed polarity and about 3-fold lower magnitudes. The results demonstrate a significant involvement of atrial repolarization during the PQ interval, and a small dispersion of atrial action potential durations. In the second part, the design and evaluation of a new optimized lead system (OACG) dedicated to atrial fibrillation is presented, based on a biophysical-model study. Considering the material constraint mentioned above, the locations of four of the six precordial electrodes were optimized while leaving the remaining five electrodes of the standard ECG system in place. The analysis was based on episodes of eleven different variants of AF simulated by a biophysical model of the atria positioned inside an inhomogeneous thorax. The optimization criterion used was derived from the singular value decomposition of the data matrices. The four new electrode positions increased the ratio of the eighth to the first singular value of the data matrices of the new configuration about five-fold compared to that of the conventional electrode positions. The OACG lead system produces a more complete view on AF compared to that of the conventional the standard 12-lead system. The third part treats the evaluation of the newly-designed OACG lead system in its application to clinical signals. Atrial fibrillation signals were recorded in patients at the nine electrode positions of 1) the standard 12-lead ECG, 2) a heuristically designed lead system, the ACG lead system, and 3) the OACG lead system. After cancellation of the ventricular signals, an information measure was derived from the singular value decomposition of the atrial signals. The resulting values obtained from the three lead systems were compared. For the limited number of recordings made available from the OACG lead system, consistently higher values of the information measure were obtained with the OACG lead system compared to the standard ECG or ACG. The ECG is clearly suboptimal in the analysis of atrial fibrillation and the OACG lead system provides a more complete view of its complex dynamics. The electric cardiac activity can be represented as the time course of a current dipole source placed inside a homogeneous thorax, the vectorcardiogram (VCG). The fourth and final topic of this thesis concerns the design of a VCG lead system committed to atrial fibrillation. Body surface potentials during atrial fibrillation were simulated by using a biophysical model of the human atria and thorax. The XYZ components of the equivalent dipole were derived from the Gabor-Nelson equations. These served as the gold standard while searching for methods to derive the vectorcardiogram from a limited number of electrode positions and their transfer coefficients. Six electrode configurations and dedicated matrices were tested using episodes of simulated atrial fibrillation and 25 different thorax models. The OACG lead system, including one electrode on the back, reduced the RMS-based relative estimation error in comparison with that of the well-known Frank lead system. The Frank lead system was found to be suboptimal for estimating the VCG during AF. Alternative electrode configurations should include at least one electrode on the back. The overall conclusion regarding these results can be recapitulated as a suboptimality of the standard 12-lead ECG system with respect to the analysis of atrial fibrillation. The key features of atrial activity are well present in body surface potentials, but appear at locations not covered by the standard lead system. While anchoring more than half of its electrodes at their conventional positions, the four new electrode positions optimized in regard to the information extraction exhibited higher performance in a biophysical-model study. The application of such an adapted lead system with its customized transfer coefficients to clinical signals promises a considerable improvement in the analysis of atrial fibrillation. The lead system with one electrode on the back of the thorax, allowing a three-dimensional capture of the complex dynamics of atrial fibrillation signals, demonstrated its utility in its application for deriving the VCG representation of the source estimation.