The appraisal and development of fractured reservoirs are challenging tasks because of the combined variations in reservoir quality and natural fracture distribution. Fracture models are built routinely to support both appraisal and development decisions for such reservoirs. One of the key objectives of these models is to generate three-dimensional (3D) fracture properties for dynamic flow simulations. In this paper, we illustrate the discrepancy between the scale of observation required to build a thorough geological understanding of the subsurface and the simplification imposed by modelling. We use a case study carried out in the context of an exploration campaign of a Cretaceous carbonate reservoir in Egypt. After describing the regional geological setting of the area, we share a series of detailed observations on open-hole logs, cores, thin-sections and borehole images. These observations are focused on reservoir architecture, fracture typology and fracture connectivity. Observations are then integrated with the regional geological context to build a conceptual fracture model and to characterize the uncertainty affecting the essential parameters of this model. The conceptual model, combined with 3D seismic data, is used to define a fracture modelling strategy. This strategy includes a drastic simplification of the conceptual model to generate 3D discrete fracture network scenarios that are calibrated using pressure communication data from the exploration wells. Another extreme simplification is then necessary to populate 3D simulation grids. In the case study presented, the key tuning parameters to obtain a dynamic match are the grid cell orientation and the width of the modelled fault damage zone.