In this paper, a feedforward neural network (NN) model is developed to predict the performance of a reverse osmosis (RO) experimental setup, which uses a FilmTec SW30 membrane. Sixty-three experimental data were generated for training and testing the network. The considered ranges of operating conditions were chosen so as to include those encountered in a large number of the worldwide brackish water and seawater RO plants. The NN was fed with three inputs: the feed pressure, temperature and salt concentration to predict the water permeate rate. The fast Levenberg-Marquardt (LM) optimization technique was employed for training the NN. The network learned the input-output mappings with accuracy for interpolation cases, but not for extrapolation. (c) 2005 Elsevier B.V. All rights reserved.