Multiphase, multicomponent equilibrium systems are encountered commonly in the chemical industry. The non-random two-liquid (NRTL) and universal quasi-chemical (UNIQUAC) models are applied widely to correlate activity coefficients in phase equilibria calculations. We demonstrated previously the predictive capability of a quantitative structure property relationship (QSPR)-generalized NRTL model (NRTL-QSPR) for vapor liquid equilibrium (VLE) property predictions of diverse binary systems. In this study, we further evaluate the prediction quality of the NRTL and UNIQUAC models as they apply to multicomponent systems. Specifically, we evaluate both the representation and prediction of vapor liquid equilibrium properties of multicomponent systems using regressed and QSPR-generalized parameters of the constituent binaries. To complete this evaluation, a database comprised of 57 ternary VLE systems and their 75 constituent binary systems was assembled. In addition, 29 ternary systems were employed to examine the efficacy of the models when at least one constituent binary is missing. For ternary systems where all three constituent binary systems were available, the NRTL-QSPR model provided overall percent absolute average deviations (%AADs) of 3.7, 0.3, 9.0, and 8.7 for pressure (P), temperature (7), mole fraction (y(1)), and K-values, respectively. In comparison, the generalized UNIQUAC-QSPR model exhibited errors of 3.5, 0.3, 8.2, and 7.9, respectively. Thus, for the systems considered, both models exhibit comparable accuracy. Further, for these systems the group-contribution method Modified UNIFAC (mUNIFAC) produced slightly better results. For systems where at least one constituent binary is missing in our database, errors for the NRTL-QSPR are 6.3, 0.5, 11.2, and 11.4 %AADs for P, T, y(1), and K-values, respectively. The comparable errors for the UNIQUAC-QSPR for this case are 8.4, 0.7, 12.9, and 12.5, respectively. For the same set of ternary systems, the mUNIFAC model produced %AADs of 20.7, 2.1, 25.7, and 26.7 for P, T, y(1), and K-values, respectively. As such, the mUNIFAC model results in three times higher errors than our generalized models. Although additional evaluation using a comprehensive database is required, the results presented in this study demonstrate the efficacy of the QSPR binary parameter generalizations, as applied to the NRTL and UNIQAC models, for VLE multicomponent property predictions. (C) 2015 Elsevier B.V. All rights reserved.