Roughness surfaces contact analysis is cutting-edge research in interface design. The 3D rough surface amplitude distribution characterized by height distribution parameters(Sq (root mean square), Ssk (skewness), Sku (kurtosis)) has a great influence on the extreme value and distribution of the interface contact stress. However, the relationship between height distribution parameters and surface maximum mises stress (σmax) is still unclear and lacks of in-depth study. Through BP neural network, global sensitivity qualitative (Morris) and quantitative (Sobol) analysis methods, the relationship between Sq, Ssk, Sku and σmaxunder different loads is studied. Based on complete polynomial and permutation combination method, the optimal correlation model between height distribution parameters and σmax was established, and particle swarm algorithm was introduced to analyze σmax extreme values under different Sq. The results show that: (1) Under different loads, the order about height distribution parameters influence on surface contact stress is: Sq> Ssk > Sku, and as the load increases, the influence of Ssk and Sku gradually decreases. (2) In different roughness surfaces, the influence of Ssk and Sku on the contact performance is significantly full of discrepancy. The research results provide reference and technical support for active design of rough surface microstructure to improved contact performance.