Modified UNIFAC (Mod. UNIFAC) is a proven model for the prediction of activity coefficients of molecules in non-ideal mixtures. However, Mod. UNIFAC is often not accurate when hydrogen bonding or strong hydrophobic interactions are present. An interesting group of molecules called polyphenols presents both types of interactions and therefore, Mod. UNIFAC predictions are not always accurate. A polyphenol-specific UNIFAC-based method (MPP-UNIFAC) has been developed in this study for its application on polyphenol-like molecules. Interaction parameters were regressed from 410 solubility data points from polyphenols in literature showing a good fit. The average error of the fit is 0.094 log units, better than the average value for Mod. UNIFAC (0.5 log units). Moreover, the model was validated against three datasets that were not used for the regression giving more accurate predictions than the Mod. UNIFAC it was developed from. (C) 2014 Elsevier B.V. All rights reserved.