A general temperature-sensitive point combination selection method based on the automatic determination of the points is proposed in this paper to solve the problem of selecting sensitive points depending on manual experience. First， the absolute mean correlation coefficients between temperature variables and thermal errors are calculated for selecting the temperature points most related to the thermal errors as sensitive points. Second， the temperature point with the largest absolute mean correlation coefficient is considered as the initial clustering center of the K-Means++ clustering algorithm， and a series of temperature-sensitive points with different numbers is selected. Subsequently， a backpropagation neural network thermal error model is established by using a series of sensitive point combinations and thermal errors as input， and the temperature-sensitive point combinations with the best prediction performance are selected based on evaluation indexes. Finally， the validity of the optimal temperature-sensitive point combination for the same error terms under different working conditions and different error terms under the same working conditions and the universality of different thermal error models are verified by employing the VMC850 CNC machine tool. The results show that the combined selection method of temperature-sensitive points proposed in this paper is suitable for experimental data under different working conditions and exhibits good versatility in different thermal error models.