Regional natural fracture networks often show variations on a scale below that captured by seismic reflection data. This variability is not considered in most reservoir models, but likely impacts uncertainties in permeability. We quantify this uncertainty using a database of 13,000 fractures in nine outcrops digitised in the carbonate Jandaíra Formation (Potiguar basin, Brazil). Distance between outcrops is on average 11 km, with a minimum of 300 m, which is comparable to the distance between wells in naturally fractured reservoirs. In between outcrops, significant variations exist in orientation, intensity, length and topology. Using discrete fracture-matrix flow models, we model the permeability of each deterministic pattern and find that small changes in geometry and topology result in permeability variations that are not captured by connectivity-based analyses such as percolation probabilities, particularly when the matrix is permeable. The permeability variations associated with subseismic-scale fracture variability are not captured in conventional stochastic models, but can be captured using deterministic outcrop models with flow through discrete fractures. The deterministic models provide a permeability range associated with subseismic fracture variability, that can