Nanofiltration can be a competitive solution to obtain drinking-water from waters with nitrate ion concentration between 50 and 150 mg L-1. The simulation of the membrane systems is necessary in order to perform optimal designs and this requires practical models to predict the effect of the composition on the nitrate ion rejection. The main intent of this study was to assess a methodology for the use of an experimental design technique called mixture design to select and analyze the optimal experiments to characterize the performance of NF membranes working with multi-ionic solutions. As complex mixtures were used to obtain experimental information, the ion interaction was taken into account. As an example, mixture design was applied to obtain polynomial models to describe nitrate rejection for feed solutions containing 100 and 300 mg (.) L-1 of nitrate ion and Na+, Ca2+ and Mg2+ cations at different proportions using the NF90 membrane. Polynomial models were adjusted using an optimal selection of experiments with adjusted r-squared values greater than 90% assuring that these models explained the rejection as a function of the factors (mol equivalents of every ion) with good agreement. The methodology can be extended to include more ions in order to obtain valid models for natural waters.