With advances in computational chemistry, phase behavior predictions without any experimental data are becoming possible. Models based on COSMO (COnductor-like Screening MOdels) are viable alternatives to group-contribution methods. These methods may be used to determine the activity coefficient of the components in any mixtures. In this paper, the model COSMO-SAC, developed by Lin and Sandler (2002), was used to predict the activity coefficients for binary mixtures (methanol + glycerol), (methanol + water) and (glycerol + water). Prior to these calculations, the software Gaussian 03 was used to perform the structure optimization of molecules of water, methanol and glycerol, in order to find which molecule conformation has lower energy. Then, sigma profiles of the components were obtained using the software MOPAC. Data ofexperimental activity coefficients (?exp) obtained from literature (Soujanya et al., 2010) were compared with calculated values (?calc) by two different methods. In the first method, the sigma profiles obtained fromsoftware MOPAC were used to calculate the activity coefficients. In the second one, they were obtained fromthe software known as JCOSMO, which uses the sigma profiles from VT-2005 database. The mean absolutedeviation between ?exp e ?calc by the two methods analyzed were calculated. The results show that the method based on COSMO not adequately predicted the activity coefficients of the mixtures analyzed for bothmethods of obtaining the sigma profile. Despite the sigma profiles generated by the software MOPAC are different from those obtained by VT-2005 database, the values of the activity coefficients obtained by both methods showed no major deviations. Furthermore, COSMO-SAC model was used to calculate phase equilibrium by minimizing the Gibbs energy of the binary mixtures. These calculations were performed in the software GAMS. It was found that, although promising, the COSMO-SAC model still need refinements.