Chemically robust membranes composed of the rubbery copolymers poly(styrene-co-butadiene) and poly(acrylonitrile-co-butadiene) along with the glassy polymer polyvinylchloride are prepared by solution casting, crosslinked with sulfur, and tested for their ability to separate benzene and cyclohexane. These polymer blends are not miscible across all compositions but careful solution casting art is used to create non-equilibrium but homogeneous blends through a process of "forced assembly." Experimental results show a typical trade-off curve; fluxes increase as selectivity decreases. Increasing the temperature from 25 to 60 degrees C results in a relatively small decrease in permeate concentration (from 93.9 to 88.3 wt.%) but to an enormous increase in flux by a factor of nearly 20 (from 5.0 to 98.9 kg mu m/m(2) h). It is demonstrated that while solubility parameters can be used to correlate the ideal swelling selectivity, they have no descriptive capability for actual membrane performance. Accordingly, a more predictive approach was investigated using group contribution methods in the form of UNIFAQ-FV. Quantitative agreement with experiment is not achieved because the equilibrium model cannot capture diffusion effects, however, the model does predict the best blend formulation (i.e. the optimal composition). Accordingly, multicomponent thermodynamic modeling can serve a limited role in providing a rational methodology for selecting blend components for specific separation or barrier purposes. (C) 2007 Elsevier B.V. All rights reserved.