In order to improve the machining precision of machine tools, an optimal model to compensate the thermal error of the spindle was explored and selected, In experinments the KVC650E vertical machining center was taken as a research object, and Multiple Linear Regression(MLR) model，Neural Network(NN) model and Support Vector Machine(SVM) model were established according to the first batch of data of the CNC center gained in autumn.By substituting the second batch of data measured in winter on the same machine and the third batch of data measured in winter on another similar machine into three kinds of models respectively,the compensation accurary of each model was calculated.The accuracy,robustness and versatility between three models was compared according to the precision variation regulation.The experiment shows that the three models have their own advantages,but SVM model ensure high precision in different environmental temperature and machine conditions,which has the best comprehensive properities.