In this paper, the dynamic behavior of reverse osmosis (RO), high-pressure (60-80 bar) membranes is investigated using an approach that combines both computational fluid dynamics (CFD) and system identification methods. Detailed CFD models of membrane systems allow predictions of the response of such systems to operating parameter changes. These models, however, are complex and require significant computing resources, which make them impractical for use in membrane system operation. In the present work, system identification theory is applied to data generated from a CFD model in order to obtain approximate transfer functions, which describe the dynamics of the system. The transfer function parameters are then related to the dynamic changes in momentum and concentration. The results show that the local concentration plays a fundamental role in the membrane performance. A feedback model, which quantifies the effect of the concentration boundary layer on the permeate flux, is finally proposed.