In order to study the infiltration characteristics of grassland soil in the super large scale landslides distribution area in the upper reaches of the Yellow River, this study selected the Xiazangtan super large scale distribution area in Jianzha County as the study area. Through experiments and numerical simulations, plant roots characteristics, soil physical properties and infiltration characteristics of naturally grazed grassland and enclosed grassland with different slope directions were compared and analyzed, and the influence of rainfall on seepage field and stability of the two grassland slopes were discussed. The results show that the highest soil moisture infiltration capacity (FIR) is found on the shady slope of the enclosed grassland (2.25), followed by the sunny slope of the enclosed grassland (1.23) and the shady slope of the naturally grazed grassland (−0.87). Correlation analysis show that soil water content, root dry weight density, total soil porosity, number of forks and root length are positively correlated with infiltration rate (P<0.05), whereas soil dry density is negatively correlated with infiltration rate (P<0.05). The results of stepwise regression analyses show that soil water content, total soil porosity, root length and number of forks are the main factors affecting soil infiltration capacity. And the ability of roots to increase soil infiltration by improving soil properties is higher than the effect of roots itself. After 60 min of simulated rainfall, the safety factors of the shady slopes of naturally grazed grassland and enclosed grassland are reduced by 29.56% and 19.63%, respectively, comparing to those before rainfall. Therefore, in this study, the roots play a crucial role in regulating soil infiltration and enhance slope stability by increasing soil water content, soil total porosity and shear strength while decreasing soil dry density. The results of this study provide theoretical evidence and practical guidance for the effective prevention and control of secondary geological disasters such as soil erosion and shallow landslide on the slope of river banks in the study area by using plant ecological measures.