Résumé : 3D Structural modeling is a major instrument in geosciences, e.g. for the assessment of groundwater and energy resources or nuclear waste underground storage. Fault net- work modeling is a particularly crucial step during this task, for faults compartmentalize rock units and play a key role in subsurface flow, whether faults are sealing barriers or drains. Whereas most structural uncertainty modeling techniques only allow for geometrical changes and keep the topology fixed, we propose a new method for creating realistic stochastic fault networks with different topologies. The idea is to combine an implicit representation of geological surfaces which provides new perspectives for handling topological changes with a stochastic binary tree to represent the spatial regions. Each node of the tree is a fault, separating the space in two fault blocks. Changes in this binary tree modify the fault relations and therefore the topology of the model.