Abstract: In order to optimize the reliability loading test of CNC machine tools and improve the consistency between the loading load and the actual cutting load, taking machining centers as the research object, the compilation of load spectra based on multi-objective optimization was investigated. Firstly, the cutting load under different cutting parameters was collected by the cutting load acquisition system, and the adaptive threshold and wavelet packet threshold were combined to reduce the data noise of the collected cutting load. Secondly, based on generalized Pareto distribution (GPD), a multi-objective optimal threshold determination method was proposed by extrapolating the cutting load extremum after noise reduction. The threshold value was selected by mean residual life (MRL) plot, and multi-objective optimization was carried out by the non-dominated sorting method. The optimal threshold value of extreme value extrapolation is obtained, and the load was reconstructed according to it. Finally, an eight-level load spectrum was compiled using the Goodman rainflow counting method. The results indicate that multi-objective optimization can effectively optimize the threshold for extreme value extrapolation and improve the accuracy of load spectra. A methodological reference for the compilation of load spectra for machining centers is provided, and provides a theoretical basis for subsequent reliability and acceleration tests.