Developing reliable EOS model for compositional simulation is important as replicating the composition phase behavior in Pressure - Temperature - Composition space is challenging. Current approaches for EOS model development are neither well defined nor justifiable in all cases. As a result, developing EOS model becomes subjective, and for same fluid, multiple EOS models can be developed with the same set of data used at regression step. All these EOS models have the same predictive capability at Pressure - Temperature points represented by data used at regression step, however, they have unknown reliability at other points in PressureTemperature-Composition space. The objective of this paper is to develop a thermodynamically consistent criterion to select most reliable EOS models out of multiple possible EOS models for a given reservoir fluid. Thermodynamic phase equilibrium condition requires that at stable equilibrium of one or more phases, total Gibbs free energy should be minimum. However, multiple EOS models because of different set of EOS parameters result in multiple values of total Gibbs free energy at Pressure-Temperature point represented by data used in the regression. This paper shows that for a given fluid, EOS model with lowest total Gibbs free energy results into most reliable phase behavior prediction. Total of 28 reservoir fluids have been considered in this study. On average 11 EOS models have been developed for a fluid by matching saturation pressure and saturation density data. For each of EOS model, Gibbs free energy of mixing and minimum miscibility pressure (for oils) and Gibbs free energy of mixing and CVD liquid saturation (for gas condensates) are calculated at the saturation pressure and reservoir temperature. Trends of Gibbs free energy of mixing and deviation in MMP and CVD liquid saturation predictions for all fluids confirms the criterion.