This study theoretically and experimentally selected a membrane for application in a perstraction process for PCB-77 removal using COSMO-RS predictions. To validate COSMO-RS, different studies of pollutants were analysed and compared using the activity coefficient computed by COSMO-RS, and the experimental reported results were in accordance with the predicted thermodynamic parameters. Then, the main components in milk were optimised using TMoleX® to screen for different polymers. The screening revealed that polychloroprene, polybutadiene, and polyoctylmethylsiloxane had the lowest activity coefficients for PCB-77, and a high selectivity for other components in milk. The results obtained by COSMO-RS were compared with experiments using nitrocellulose, polydimethylsiloxane, polyether sulfones, and polystyrene, where the simulated values were validated. Finally, analysis of the membrane structures revealed that the best membrane to be applied in perstraction is polyoctylmethylsiloxane.