The lack of 3D precise data in fractured reservoirs makes it difficult to efficiently characterize and model them. In this article, a new approach is proposed to integrate all available data and their uncertainties into the reservoir model, and to simulate a Discrete Fracture Network (DFN) in order to assess the reservoir permeability. From all available data, 3D distributions of probability laws are associated with each of the major parameters describing the fracture network, such as fracture density, orientation, size and aperture. Fracturation patterns highly depend on stress history. This stress field can be evaluated through a 3D balanced unfolding of the reservoir that gives a strain tensor at each point of the model. Then, assumptions on the values of mechanical rock properties (cohesion, Young's modulus, ...) and on their uncertainty, represented as a probability law, allow to compute a 3D distribution of probability of fracturation. Moreover, using a non-linear multi-regression, this probability is combined with other fracture guides, such as seismic attributes and layer thickness, to obtain a fracture density analogue that reproduces as well as possible observed well data. The mismatch between the analogue and the well data reflects the quality of the fracturation model. Mechanical analysis and well data also provide 3D distributions of fracture size and orientation probability law, which are used for generating DFNs using a heterogeneous Poisson point process. Methods for computing an equivalent fracture permeability are finally discussed.