Soil erosion has become a serious ecological problem in many catchments. Soil erodibility K factor can be estimated based on a series of soil properties, however, the identification of dominant soil properties that affect K factor prediction at different soil types has been little concerned. In this study, 3 soil profiles from the Jiulongjiang River Catchment (JRC) of granite region in Fujian province and 18 soil profiles from the Chenqi Catchment (CC) of karst region in Guizhou province were selected. Soil properties, including soil particle size distribution, soil organic carbon (SOC) and soil organic nitrogen (SON) content, and soil pH, were determined, and the K factors were estimated in the erosion productivity impact calculator (EPIC) model. The soils in the granite region were characteristic for coarse texture, low SOC and SON, and strong acidity compared with limestone soils. Although the K factors in both regions ranged from 0.009 to 0.018, they were overestimated in limestone soils due to frequent soil aggregation, which enhanced soil permeability, hence reduced soil erodibility. The results of principal component analysis (PCA) and structural equation model (SEM) showed that (1) K factor estimation in the soils of the granite region mainly depended on soil texture, of which silt was the most important factor; (2) while K factor in limestone soils was mainly controlled by soil organic matter (SOM) content, other soil properties, including soil pH, clay and silt contents, could indirectly affect prediction of K factor by affecting SOM accumulation.