Proper understanding and mapping of a geological domain calls for a good characterization of fault slip. Several fault displacements can be proposed to explain a given geometry of the three-dimensional geological model. Most approaches ignore this, and assume a fixed direction of fault slip, for instance along the main dipping line. Instead, we propose to generate several fault displacement models that all conform to available geometrical data and a given structural deformation style. The method uses sequential Monte-Carlo sampling to simulate fault displacement vectors on a geological horizon; balanced restoration is run to compute the likelihood of the fault net slip. Acceptable models are selected using the Metropolis-Hastings algorithm. This procedure allows selecting the best model of fault displacements given the retained structural style. Alternatively, the fault throw configurations obtained can be input in a global uncertainty assessment workflow. To assess the uncertainty of fault properties (e.g., shale-gouge ratio), and of stratigraphic transforms such as the geo-chronological parameterization. The sampling method is demonstrated on a complex, heavily-faulted stratigraphic model. A discussion on the principle of parsimony and deterministic versus stochastic modeling is included.