Nonlinear analysis has emerged as a novel methodology for the processing of physiological time series over the past few decades because biological systems and processes are inherently complex, nonlinear, and non-stationary. In this sense, cardiac cells and the time course of atrial electrophysiological properties have shown a far-from-linear behavior during atrial fibrillation (AF). This work reviews the main approaches of nonlinear analysis (chaos theory, information content quantification, irregularity measures, and geometric features) in the study of AF. The application of these indices both to surface electrocardiographic recordings and invasive atrial electrograms reveals useful clinical information in a complementary way to traditional linear techniques, such as spectral analysis. Indeed, atrial activity signal analysis through nonlinear indices provides more accurate estimates of the temporal, spatial, and spatio-temporal organization of AF. Given that AF organization has been associated with the number of simultaneous wavelets wandering throughout the atrial tissue during arrhythmia, its robust estimation can help improve treatments and allow more appropriate decisions on AF management.