The compensation of thermal deformation is the most significant for the accuracy of a machine tool. This study proposes an approach based on genetic algorithms (GA) to build the dynamic model of the prediction for thermal deformation of a machine tool. GA is used to optimize the prediction accuracy by using appropriate number and locations of temperature sensors, the model order and the time delay between temperatures and thermal deformation. The compared results show that the proposed approach can improve the accuracy of prediction results and better than other methods.