Reservoir production curves provide significant information about reservoir compartmentalization and connectivity, but are most often used at late stages of reservoir modeling workflows, leading to possible inconsistencies between static and dynamic reservoir models. In this paper, we propose to use this information during structural modeling to reduce fault-related uncertainties induced by subsurface imaging and interpretation ambiguities. We first generate stochastic fault networks from prior information about faults hierarchy, size, orientation, roughness and localization. Then, the results are discretized into a set of nodes and pipes, called a pipe network, which represents the control volumes connectivity graph. This discretization data structure accurately represents fault surfaces and fault connections while limiting the number of nodes to represent the domain of interest as a connectivity graph that can be fed to a flow simulator. In a third step, we select models that best match the production curves. The proposed method is flexible enough to assess high uncertainties that may require to add or remove faults, or to perturb the fault connectivity, without need for simplification of the fault network into a corner-point reservoir grid. The method is applied to a synthetic example of high uncertainties relative to a poorly imaged 3D fault zone.