Atrial fibrillation (AF) is a common and increasingly prevalent condition in the western society. During AF, the AV-node controls ventricular response to the rapid atrial impulses. However, current research indicates that the individual variability in AV-nodal function is large. Thus, characterization of the AV-node is an important step in determining the optimal form of treatment on an individual basis. Here we employ a multilevel modeling approach, comparing a previously presented statistical model with a novel detailed network model of the AV-nodal function during AF. We demonstrate that both models can be fitted to generate output that closely resembles clinical ECG data, and that estimated parameters in the less complex model corresponds to limited ranges of parameters in the more complex model.