In the present work, a knowledge model is used to investigate the possibility of predicting the separation of ionic mixtures. In the model, the transfer through the membrane is based upon the classical vision coupling the Donnan, steric and dielectric exclusions at the interfaces with the extended Nernst-Planck transport equation within the pores. The prediction of selectivity between ions requires the knowledge of the interfacial partitioning which results of three contributions, i.e. steric, electric and dielectric. Each contribution can be assessed by means of a parameter describing the exclusion phenomena. The mean pore radius (r(p)) and the membrane charge density (X(d)) which govern respectively steric and electric exclusions are usually experimentally identified by microscopy or neutral solute rejection and zetametry. The dielectric contribution through the dielectric constant of the solution within the pores (epsilon(p)) is usually fitted from rejection curves. In this paper, an experimental method based upon the membrane potential measurements is also implemented to assess epsilon(p)-value. The identified values are compared with those obtained by fitting experimental ternary mixtures. The agreement found between these two sets of values indicates that the theory of ionic transfer seems to be coherent and let foresee the possibility of a fully predictive model. (C) 2011 Elsevier B.V. All rights reserved.