The nanofiltration of binary aqueous solutions of glucose, sucrose and sodium sulfate was investigated using thin-film composite polyamide membranes with different molecular weight cut-off s. The NF experiments, in total recycle mode, were performed in a plate-and-frame module Lab 20 (AlfaLaval), at 22 degrees C and with a flowrate of 8.2L/min, using the membranes NF90, NF200 and NF270 from FilmTec (Dow Chemical), for transmembrane pressures between I and 6 MPa and with aqueous solutions with osmotic pressures of between 0.5 and 3.0 MPa. The permeate flux was predicted by the osmotic pressure model, using the membrane hydraulic resistance and the solution viscosity inside the membrane pores, and computing the concentration polarization with recourse to a mass-transfer correlation specific for the plate-and-frame module used. The flux predictions, using the pure water viscosity, agree reasonably with the experimental data only for low transmembrane pressures and with the most diluted solutions. For higher transmembrane pressures and for higher solute concentration the predicted fluxes can be as far as 2.5, 4.1 and 9.6 times higher than the experimental one, for the aqueous solutions of Na2SO4, glucose and sucrose, respectively. These deviations are strongly reduced when the pure water viscosity is replaced by the solution viscosity adjacent to the membrane. In this case, the maximum deviation between predictions and experiments occurs also for higher transmembrane pressures and for higher solute concentration, but the maximum ratio between predicted values and the experiments were reduced now to 1.8, 2.1 and 2.9, for the aqueous solutions of Na2SO4, glucose and sucrose, respectively. Even using the solution viscosity adjacent to the membrane, and for the systems investigated, the osmotic pressure model must used with caution for design purposes because it may over predict the permeate flux by a factor of about 2 when the solute concentration is high. (C) 2009 Elsevier B.V. All rights reserved.