To optimise recovery in naturally fractured reservoirs, the field-scale distribution of fracture properties must be understood and quantified. We present a method to systematically predict the spatial distribution of natural fractures related to faulting and their effect on flow simulations. This approach yields field-scale models for the geometry and permeability of connected fracture networks. These are calibrated by geological, well test and field production data to constrain the distributions of fractures within the inter-well space. First, we calculate the stress distribution at the time of fracturing using the present-day structural reservoir geometry. This calculation is based on a geomechanical model of rock deformation that represents faults as frictionless surfaces within an isotropic homogeneous linear elastic medium. Second, the calculated stress field is used to govern the simulated growth of fracture networks. Finally, the fractures are upscaled dynamically by simulating flow through the discrete fracture network per grid block, enabling field-scale multi-phase reservoir simulation. Uncertainties associated with these predictions are considerably reduced as the model is constrained and validated by seismic, borehole, well test and production data. This approach is able to predict physically and geologically realistic fracture networks. Its successful application to outcrops and reservoirs demonstrates that there is a high degree of predictability in the properties of natural fracture networks. In cases of limited data, field-wide heterogeneity in fracture permeability can be modelled without the need for field-wide well coverage.