The discrimination among atrial rhythms is obtained through the analysis of linear and non-linear dynamics of atria electrograms and local atrial period (LAP) series. Linear and non-linear degree of coupling between atrial electrograms have been assessed using cross-correlation (CCI) and synchronization indexes. Dynamics of LAP series were investigated using level of predictability (LP) and regularity indexes. Atrial fibrillation (AF) episodes, classified according to Wells' criteria, were investigated. We found that metrics obtained from LAP series provide the best performance in the classification of AF rhythms. The LP index misclassified only 12 AF episodes over 87. Synchronization index was the most performing metric in the discrimination between organized and not-organized fibrillation: detection of organized rhythm; sensitivity (SE) > 95%; positive predictability (+P) > 95%. LP (SE: 89%; +P: 89%) and CCI (SE: 87%; +P: 83%) provide slightly lower performances.