Thermal error is one of the main factors that influence the machining accuracy of computer numerical control (CNC) machine tools. It is usually reduced by thermal error compensation. Temperature field monitoring and key temperature measurement point (TMP) selection are the bases of thermal error modeling and compensation for CNC machine tools. Compared with small- and medium-sized CNC machine tools, heavy-duty CNC machine tools require the use of more temperature sensors to measure their temperature comprehensively because of their larger size and more complex heat sources. However, the presence of many TMPs counteracts the movement of CNC machine tools due to sensor cables, and too many temperature variables may adversely influence thermal error modeling. Novel temperature sensors based on fiber Bragg grating (FBG) are developed in this study. A total of 128 FBG temperature sensors that are connected in series through a thin optical fiber are mounted on a heavy-duty CNC machine tool to monitor its temperature field. Key TMPs are selected using these large-scale FBG temperature sensors by using the density-based spatial clustering of applications with noise algorithm to reduce the calculation workload and avoid problems in the coupling of TMPs for thermal error modeling. Back propagation neural network thermal error prediction models are established to verify the performance of the proposed TMP selection method. Results show that the number of TMPs is reduced from 128 to 5, and the developed model demonstrates good prediction effects and strong robustness under different working conditions of the heavy-duty CNC machine tool.