A thermal error modeling method of NC machine tools based on deep learning method was proposed in order to improve the prediction ability of thermal error model. Fuzzy clustering method and grey relationship analysis method were used to select the sensitive points of temperature variables and the stacked automatic encoder (SAE) network was used to extract the features of the temperature variables from the selected input samples to construct the feature sets. Then, genetic optimization algorithm (GA) was used to optimize BP neural network parameters so as to propose a thermal error modeling method based on SAE-GA-BP neural network for NC machine tools. Taking a large gantry fivesided machining center as the experimental object, the spindle thermal error of the large gantry fivesided machining center was studied and selected as the main error source to achieve compensation in the machining process. The deep learning model of main shaft thermal error was compared with the multiple regression model. The experimental results showed that the proposed modeling method was better than the traditional multiple regression model in prediction accuracy of the thermal error of NC machine tools, which verified the feasibility and effectiveness of the proposed thermal error modeling method.