Abstract: The influence of meso-parameters on macro-mechanical response of loess in triaxial test is studied from meso-perspective. Based on the PFC^3D software and the results of the indoor triaxial test to determine the value range of the basic mesoscopic parameters of the numerical simulation, particle flow simulation of triaxial samples is performed under the confining pressures 50 kPa, 100 kPa, 150 kPa, and 200 kPa. By changing the influence degree of mesoscopic parameters on the macroscopic mechanical behavior of loess numerical samples and orthogonal test design, the sensitivity analysis of friction coefficient, porosity, particle stiffness ratio k_n/k_s and particle size distribution in the process of loess triaxial numerical simulation is carried out, and the relationship between soil mesoscopic parameters and macroscopic mechanics is established. The results show that the peak strength and residual strength of the numerical samples under different confining pressures are positively correlated with the friction coefficient, and negatively correlated with the particle stiffness ratio and porosity. The influence of particle size distribution on it is less than 2%. Under low confining pressure(50 kPa), porosity has the greatest impact on the peak shear strength and residual strength of numerical samples, followed by particle stiffness ratio k_n/k_s, and under high confining pressure(200 kPa), the influence of friction coefficient on the strength of numerical sample is more significant than that of particle stiffness ratio. The meso-parameters that have the greatest impact on the initial linear elastic modulus and strain softening characteristics of the numerical sample are porosity, followed by the stiffness ratio k_n/k_s, and the particle size distribution is the smallest. The result of orthogonal test under 200 kPa confining pressure is consistent with the result of numerical analysis. By changing the value range of meso-parameters, the influence on the macro-mechanical properties of the loess triaxial test is systematically analyzed, which provides a reference for the future use of discrete element software in loess laboratory test research.