Atrial fibrillation is commonly described by irregular electrical excitations circulating in the atria. Developing methods to characterise and classify various kinds of atrial fibrillation is an important task, since it allows one to improve the diagnosis and to support the choice of suitable therapies. While most methods use invasive intraatrial measurements by catheter electrodes, one can obtain valuable information also from the noninvasive surface ECG by studying the statistics of ventricular impulses. Such approach, however, requires a good understanding of the conduction of impulses through the atrio-ventricular node. Based on a simple conduction model we study how the distribution of ventricular interbeat intervals is affected by the atrial fibrillation. Due to the nonlinear properties of the AV node, a surprisingly rich pattern of behaviours is found, even if the fibrillation is characterised in terms of a Poisson process with only one characteristic fibrillation rat