In order to improve the prediction accuracy of the thermal error of the CNC lathe spindle,a CNC lathe spindle was taken as the research object,and a thermal error modeling method based on wavelet neural network (WNN) was proposed.The K-means++ clustering and correlation analysis theory were used to determine the temperature measuring points from 10 to 2.Aimed at the problem that the WNN was sensitive to the initial value,the bat algorithm (BA) was used to calculate the absolute value of the difference between the predicted output value and the experimental measurement value as the individual fitness function,and the position vector of the bat individual was mapped to the initial connection weight,scale factor and translation factor of the WNN,so the initial value of the WNN was optimized.The optimized WNN was used to establish the thermal error prediction model of the lathe spindle,and then it was compared with the un-optimized WNN and BP neural network prediction model.The results show that the BA-WNN has higher prediction accuracy,less residual and stronger prediction ability for the axial thermal error of the spindle.