Lactose retention was predicted by means of the Donnan Steric Pardoning model (DSPM) and the Kedem-Spiegler model (KSM). Retention and permeate flux were measured using single lactose solutions and ultrafiltered whey (UF-whey) as feed solutions. The experiments were performed at constant pH with NF200 and desal 5 DL nanofiltration membranes supplied by Dow-Chemical (USA) and GE-Osmonics (USA), respectively. The two models were able to predict reasonably well the pattern and tendencies of lactose retention as a function of permeate flux when either a lactose solution or UF-whey was used as feed. The best predictions were obtained with the KSM. This model was observed to be able to predict lactose retention in the presence or absence of ions without providing any information about the membrane structure. However, the DSPM provides information about the charge, the thickness-porosity ratio and the pore size of the membranes. This contributes to a better understanding of the transport mechanism in nanofiltration processes. The presence of salts increased the values of the fitted mean pore radius and the fitted thickness/porosity ratio of the membranes. For the two membranes tested lactose retentions were lower in the case of UF-whey. This was explained taking into account the effect of salts on membrane performance. (c) 2007 Elsevier B.V. All rights reserved.