In order to reduce the influence of thermal error of CNC machine tool on machining accuracy and to realize the compensation control for thermal error, a thermal error modeling method for CNC machine tool based on least squares support vector machine(LSSVM) optimized by genetic algorithm(GA) was proposed. The genetic algorithm was used to optimize the selection of LSSVM penalty factorCand the kernel function parameter σ2, a thermal error GA-LSSVM model for a horizontal machining center spindle was constructed. According to this model, the comparison curve of the simulated value and measured value of the thermal error was obtained. Through analysis, it was found that the GA-LSSVM model had good performance, small residual and high prediction accuracy. The thermal error LSSVM model and traditional BP model were established and they were compared with the GA-LSSVM model. The results show that the absolute residualδand mean square error (MSE) of GA-LSSVM model are both the minimum, and the model determination coefficient R2 is the maximum. It proves the effectiveness of the GA-LSSVM modeling method