Atrial fibrillation is the most common cardiac rhythm disorder encountered in clinical practice, often leading to severe complications such as heart failure and stroke. This arrhythmia, increasing in prevalence with age, already affects several millions of people in the United States, with a rising occurrence of the disease during the past two decades. In spite of these warning signals, atrial fibrillation is still difficult to treat, because basic mechanisms of the arrhythmia remain poorly understood and current treatments are therefore based on empirical considerations. The future of therapeutic solutions for the treatment of complex diseases such as atrial fibrillation relies on a strong collaboration between medicine, biology and engineering. Only through such synergies will efficient monitoring, diagnostic and therapeutic devices be created. The goal of the present thesis was to adopt this multidisciplinary approach, and develop new strategies for atrial fibrillation therapy using both computer modeling and advanced signal processing methods. Biophysical modeling is a practical and ethically interesting approach to develop innovative therapies, since physiological phenomena of interest are reproduced numerically and the resulting framework is then used with full repeatability to explore mechanisms and test treatments. A model of the human atria, that was developed in our group, was used to simulate atrial fibrillation and perform mechanistic and therapeutic investigations. In a first study, computer simulations were used to observe spontaneous terminations of two models of atrial fibrillation corresponding to different developmental stages of the arrhythmia. Dynamical parameters were observed during several seconds prior to termination in order to describe the underlying mechanisms of this natural phenomenon, showing that different levels of fibrillation complexity led to different termination patterns. The mechanisms highlighted by the study were successfully compared to those described in the existing literature and could suggest interesting guidelines to better investigate spontaneous terminations of atrial fibrillation in experimental and clinical settings. Moreover, a more precise understanding of the natural extinction of atrial fibrillation will certainly be crucial for future therapy developments. The potential of rapid low-energy pacing for artificially terminating atrial fibrillation was also thoroughly investigated. First, the possibility to entrain and thereby control fibrillating atrial activity by rapid pacing was studied in a systematic manner. Results showed that optimized pacing parameters provided sustained entrainment of electrical activity, although total extinction of atrial fibrillation was never observed. The ability to control atrial activity by pacing was also shown to depend on specific properties of the atrial tissue, showing that patients with atrial fibrillation may not all respond in the same way to pacing treatments. Finally, this study suggested different guidelines for the development of pace-termination algorithms for atrial fibrillation. Based on these results, a new pacing sequence for the automatic termination of atrial fibrillation was designed, implemented and tested in the biophysical model. The pacing protocol comprised two distinct phases involving a succession of rapid and slow pacing stimulations. The results of the tests suggest that this pacing scheme could represent an alternative to current treatments of atrial fibrillation, and could easily be implemented in patients who already have an indication for pacing. Advanced signal processing techniques were also used in this thesis to analyze real cardiac signals and develop new diagnosis tools. Multivariate spectral analysis and complexity measures were combined to develop an automatic method able to describe subtle changes in atrial fibrillation organization as measured by non-invasive ECG recordings. Accurate discrimination between persistent and permanent AF was shown possible, and potential applications in clinical settings to optimize patient management were demonstrated. Collectively, the results of this thesis show that major public health issues such as atrial fibrillation can strongly benefit from the contribution of biomedical engineering. The modeling and signal processing approaches used in the present dissertation proved effective and promising, and synergies between clinicians and scientists will definitely be at the basis of future therapies.