Lower critical solution temperature (LCST) and upper critical solution temperature (UCST) are key thermodynamic properties of binary polymer solutions. In this work, quantitative structure property relationship (QSPR) modeling was employed to predict θ(LCST) and θ(UCST) (LCST and UCST at the limit of infinite molar mass). Based on a series of topological norm descriptors and quantum chemical norm descriptors derived exclusively from the chemical structures of polymers and solvents, four linear topological and spatial LCST-QSPR and UCST-QSPR models were developed. The accuracy, robustness, and predictability of the proposed models were evaluated in detail by various statistical parameters (e.g., R2, MAE, MRE, and RMSE) and validation approaches (e.g., leave-one-out cross validation and Y-randomized validation). Various validation techniques and statistical indicators reveal that the spatial LCST-QSPR and UCST-QSPR models established by adding quantum chemical norm descriptors show better performances. Desirable agreements (Rtraining2 = 0.9423) between calculated and experimental θ(LCST) of 118 training set polymer solutions can be found in the spatial LCST-QSPR model as demonstrated by MRE of 2.65 %. Meanwhile, the spatial UCST-QSPR model shows the performances with MRE of 8.03 % and Rtraining2 of 0.8826 of 87 training set polymer solutions. The comparative results with those of different models from literatures further confirm the advantages of the as-developed models. The presented QSPR models are expected for rapid and accurate prediction of the θ(LCST) and θ(UCST) values of various binary polymer solutions.