Abstract: Surface integrity is a key factor influencing the contact fatigue performance of gears. Exploring the quantitative correlation between surface integrity parameters and the contact fatigue limit of gears is crucial for achieving active design based on surface integrity parameters and enhancing gear fatigue performance. To explore the quantitative correlation between surface integrity parameters and the contact fatigue limit, characterization of surface integrity and contact fatigue limit testing for 18CrNiMo7-6 and 16Cr3NiWMoVNbE carburized gears under different process states including grinding, shot peening, dual shot peening, and barrel finishing after shot peening. The key surface integrity parameters influencing the contact fatigue limit of gears were analyzed using Pearson correlation coefficient method, and the influence weight of feature parameters on the contact fatigue limit of gears was explored based on CatBoost algorithm. The prediction formula for the contact fatigue limit of gears based on surface integrity parameters was derived using multiple linear regression method. The results show that in the dual shot peening state, the maximum subsurface residual stress amplitude of 18CrNiMo7-6 gears can exceed 1250 MPa, with a surface hardness reaching 685 HV, and the contact fatigue limit can be increased by up to 19.6% compared to the carburized grinding state. Furthermore, after being treated with composite surface strengthening processes such as barrel finishing after dual shot peening, the residual compressive stress amplitude on the surface of 18CrNiMo7-6 gears can reach 1150 MPa, and the contact fatigue limit is increased by more than 20% compared to the carburized grinding state. Based on the gear materials and manufacturing processes discussed in this paper, the surface integrity parameters most relevant to the contact fatigue limit are surface hardness, maximum residual compressive stress, surface residual compressive stress, and surface roughness, with contributions of 39.5%, 24.7%, 23.5%, and 12.3%, respectively. The proposed gear contact fatigue limit prediction formula based on surface integrity has an average prediction error of only 2.9% compared to experimental results, meeting engineering application requirements.