The objective of this work was to improve the accuracy of group contribution models for prediction of solvent activities in polymer solutions by revising UNIFAC group interaction parameters using a wide range of vapor-liquid equilibrium (VLE) data of solvent-polymer systems. The group contribution models considered in this work were UNUFAC-FV, Entropic-FV, GK-FV and UNIFAC-ZM models. A total of 142 systems that consisted of 16 polymers and 36 solvents containing a large variety of solvent-polymer systems ranging from non-polar to polar substances were considered to optimize 46 pairs of group interaction parameters. Data considered were split up into systems containing alkane and cycloalkane, aromatic, and polar solvents. For athermal systems, the UNIFAC-FV model gave the best results. Therefore, the model was used in optimizing the group parameters. Revised group interaction parameters were found to improve the reliability of VLE predictions in solvent-polymer systems. A significant improvement of prediction results was achieved by UNIFAC-FV model from 20.0 to 10.8% absolute average deviation (AAD) in solvent activities for systems containing polar solvents and from 16.7 to 10.9% AAD for all systems. The prediction results of GK-FV and UNIFAC-ZM models were also improved.