Current group-contribution methods such as ASOG and UNIFAC are widely used for approximate estimation of mixture behavior but unable to distinguish between isomers. Atoms in Molecules (AIM) theory can solve these problems by using quantum mechanics and computational chemistry to compute atomic contributions to molecular properties and to intermolecular interactions. Rigorously defined properties available through AIM theory and new functional group definitions are used for the UNIFAC model to predict the behavior of various mixtures. Results are presented for various mixtures with nine regressed global parameters to optimize model's predictive capability. The results are also compared to analogous results for the Knox model.