Polycyclic aromatic compounds (PACs) are common environmental contaminants associated with oil spills and the incomplete combustion of organic materials. PAC solubility in water is a fundamental property for environmental studies, and modeling of these data improve the process of environmental risk assessment. In this study, seven models (UNIQUAC, local surface Guggenheim, NRTL, regular solution, Wilson, Van Laar, and a modified Van Laar model) for correlations and prediction of aqueous solubilities of 22 PACs were evaluated. The results using models based on Guggenheim's method showed that the local surface Guggenheim model provided a better correlation than the UNIQUAC model. For the systems studied, the best correlations were obtained with NRTL, Van Laar, and modified Van Laar models with mean deviations of 17.1, 14.3, and 14.5%, respectively. The predicted solubilities using NRTL and modified Van Laar models provided mean deviations of 44.1 and 47.1%, respectively. The sensitivity analysis showed that the correlations using the NRTL model are slightly influenced by variations up to 20% of the triple-point temperature and molar enthalpy of fusion of the solute.