The structural interpretation of 2D seismic lines and their correlation to deduce a 3D structure, e.g. a fault network, is a hard task for geoscientists. Interpretations may introduce a prior geological knowledge bias and/or human bias (Bond et al., 2007). Consequently, decisions based on one single deterministic model of the subsurface may be highly risky. The goal of our study is to show that the 3D organization of a fault network, especially fault connections, is uncertain and to investigate how these uncertainties evolve depending on the quantity of input data. For this, we realized stochastic simulations of fault networks using a new method that allows both geometry and fault connection changes from statistical parameters about fault geometry and fault traces interpreted on 2D seismic lines. Simulations have been performed for two scenarios in which the quantity of input data was different, and generate a set of possible models whose diversity decreases as more data are added.