To diminish the thermal error and to enhance the machining precision of CNC machines, thermal error compensation is an effective way. Taking GMC4000H/2 machining center as research object, this paper focuses on the thermal compensation skills of machine tools. Firstly, detection method of thermal deformation error on machine tool was introduced. And then, this paper established compensation models between measured temperature parameters and thermal error parameters using PSO-BP neural network. Finally, a thermal compensation controller that was used to implement error compensation was developed. The thermal compensation controller hardware and simulation result were given. The simulation experimentation showed that the PSOBP neural network can enhance precision from 93?m to 13?m, and also the model had good performance of compensation and fitting.