As hydrodynamics conditions govern membrane fouling, simulations with a Computational Fluid Dynamics software were run to establish optimal design of membrane network. First, simulation through one cylinder was used to calibrate model by comparing the separation angle of experiments executed by Sucker & Brauer (1975) and the separation angle of simulations. Moreover, streamlines profile were compared for different Reynolds numbers. Intermediate or turbulent flow (Red = 600) imposed the choice of turbulence models. It is proved that the standard model, k-epsilon, is not well adapted to simulate the flow around a curved body. The SST-k-omega model resulted in better simulations. Geometry and boundary conditions were taken as common used (Newman and Dirichlet). Once the model was calibrated, simulations were run with a bundle of capillary fibers (membrane network). Velocity, velocity gradient, and friction coefficient were used to define optimal design according to experimental results