Fault representation and scaling in flow models are examined with respect to fault zone properties, the accuracy with which they can be determined, and how these variables and fault geometries can be incorporated realistically in to flow models. Outcrop data show that fault displacement/thickness ratios and permeability vary widely. For simple single fault models, results for numerical models are compared with analytical and statistical methods. Representation of a fault as a transmissibility surface conflates the effects of four variables--fault zone thickness and permeability, grid-block size and matrix (host-rock) cell permeability. Random spatial variation of transmissibility factor values is well represented by a uniform transmissibility factor which is the arithmetic mean of the values representing log-normally distributed permeability and thicknesses. Realistic ranges of fault zone thicknesses can be represented without grid-block refinement by an upscaling method based on simple transformation of transmissibility factor curves derived from a range of coarse grid-block models. Sub-seismic faults have significant effects on effective permeability of model volumes at kilometre scales only when the faults are assigned a permeability less than c. 0.001 of the matrix permeability.