Benggang, a compounded erosional landform influenced by geological tectonics, geomorphological conditions, meteorological hydrology, vegetation, and human activities, is widely distributed in southern China. The formation and development of Benggang severely restricted local agricultural production and socio-economy. Adjusting land use structure and optimizing landscape pattern are crucial ways to prevent Benggang erosion in red soil region of southern China. How to investigate the relationship between landscape pattern and Bengggang erosion at patch-class level, which is of great significance for controlling Benggang and improving eco-environmental benefits. Herein, a case study with regard to investigate the impact of landscape indices of various patch types on the area and density of Benggang in Anxi County, Fujian Province in southern China. The study area was divided into 53 sub-watersheds based on DEM data using the ArcGIS hydrological analysis tool. The area and density of Benggang were taken as the response variables, and the landscape pattern indices were the explanatory variables. Effects of landscape pattern indices on the response variables were determined by redundancy analysis (RDA) and path analysis at patch-class level. The results indicated that the landscape metrics affecting Benggang erosion were mainly determined by land use types. Specifically, the landscape metrics of woodland had the highest interpretation (63.82%), followed by farmland (49.19%) and land used for building (42.01%). The total class area of farmland, clumpiness of woodland, landscape shape index of land used for building, and patch cohesion index of woodland contributed to 51.5%, 26.9%, 9.2%, and 6.1% of Benggang erosion, respectively. Meanwhile, the integrated impacts of these four variables could explain 66.2% of Benggang variance. Furthermore, the path analysis results showed that the total class area of farmland, patch cohesion index of woodland, edge density of woodland, percentage of landscape of woodland, and edge density of land used for building could jointly explain 71% of the variation in Benggang area and 85% of the Benggang density. Among these five landscape patterns, the total class area of farmland and area percent of landscape of woodland were the most critical factors to Benggang area and density, and the path coefficients were 0.499, -0.440 and 0.608, -0.342, respectively. In general, a scientific spatial layout (such as farmland, woodland, and land used for building) benefits to an improved ecological effect of landscape types, significantly influencing the control of Benggang. This study facilitates scientific planning of land utilization and provides effective measures for Benggang control in red soil region of southern China.