Abstract Faults significantly influence fluid flow in reservoirs. In standard reservoir flow-simulator grids, faults are represented as surfaces or planes. However, outcrop studies show that faults often must be regarded as volumetric elements. Inside such fault zones, the facies characteristics differ significantly from those in the rest of the reservoir. In a fault facies reservoir model faults are represented as volumes populated with facies with properties derived from their origin and faulting history. In this paper, we compare fluid flow performance of a fault facies model and a conventional fault model. The uncertainties attached to the fault zone properties and architecture included in the fault facies model produce a straightforward effect on the range of simulation outcomes and uncertainty of production parameters. In the conventional model, similar effects can only be reproduced ad hoc using poorly determined random fudge factors. We also look at the effect of upscaling the fault zone. Results show that the flow properties in the fault facies model differ from the conventional model with regard to both water cut and total oil production. As expected, upscaling may introduce a significant bias in the cumulative oil production.