During the machining process,the machine tool will produce errors due to thermal deformation,which will seriously affect the machining accuracy. Reducing thermal error of machining process is an effective way to improve machining accuracy.In the process of reducing thermal errors,determining key temperature measurement points can not only improve the calculation efficiency,but also avoid the problem of collinearity between temperature data and improve the prediction accuracy of the thermal error model. A method for selecting temperature measurement points based on improved fuzzy clustering and maximum information coefficient (MIC) was proposed,and the temperature measurement points were classified by improved fuzzy clustering;the key temperature measurement points in each type of temperature data were selected according to the MIC method;the BP neural network was used to model the thermal error. The results show that compared with the traditional method of selecting temperature measurement points,the accuracy of the thermal error model improved by using this method is higher.