Ethanol dehydration systems are modeled and optimized using MINLP. The systems consist of a distillation column for approaching the ethanol/water azeotrope and a pervaporation system for producing pure ethanol. Rigorous modeling equations are employed for modeling both the column and the subsystem of membrane modules. Quadratic and exponential regression, as well as metric and linear interpolation are studied for approximating the integral membrane model. A method for radically decreasing the multiplicity of equivalent structures covered by the superstructure is suggested and applied. A successive refinement method with non-increasing number of binary variables is developed and successfully applied.