【Objective】Soil erodibility K is an international index of soil susceptibility to erosion, and can be used as an important quantitative parameter in evaluating soil erodibility. The Yanhe Valley is located on the Loess Plateau, where soil erosion is very severe, and also very severe in soil erosion. In this case, it is particularly important to conduct research on soil erosion models for this region. In recent years, in studies on soil erodibility of loess, soil erodibility factor K is often used as an index for evaluation of soil erosion. Though certain progress has been made in the research on using the formula method to assess soil erosion factor K in the loess area, it is still infeasible to go on doing researches on estimating K values in some parts of the Loess Plateau due to limitation of data availability and inconsistency between standard plot and observation plot. Besides, the reliability of the formula method still need to be validated. So, it is necessary to design an equation that is workable for estimating soil erodibility K even when inadequate data of soil physical and chemical properties are available. The purpose of this study is to pick out of the three methods currently available for estimating soil erodibility K one that fits the special situation of the river valley.【Method】In this study, comparison was performed between the three methods, i.e. Torri.D model, EPIC model, and Shirazi formula in applicability to estimation of K for the nine catchments of the Yanhe Valley. Collection analysis and Model-based estimation methods were used to process and analyze the data and compare predicted K with measured K, so as to screen out the most suitable one.【Result】Results show that the contents of soil organic carbon, clay and silt gradually increased from north to south with the increasing vegetation coverage. In terms of mean weight diameter (D_MW), the three types of vegetation in the valley followed an order of forest > forest-steppe > steppe, and D_MW was positively related to the K predicted with the EPIC model and Shirazi formula method, but negatively to that with the Torri.D model, which means that soil aggregate increased in stability and the soil in erosion resistance as the vegetation turned from steppe to forest-steppe to forest. The three predicted Ks displayed an order of K_Torri.D> K_EPIC > K_shirazi K_Torri.D varied in the range of 0.068 ~ 0.1475, higher than the measured one (0.0312~0.0796). Compared with the other two, Torri.D model was the lowest in uncertainly, with mean absolute error (MAE), mean relative error (MRE), root mean square error (RMSE) close to 0, and dilution of precision (A_f) close to 1, suggesting that Torri.D model is more suitable than the other two for use to evaluate soil erosion susceptibility and calculate soil loss.【Conclusion】To sum up, all the findings described above indicate that Torri.D model can be used to soil erosion susceptibility and predict soil loss of a region even when data of the region are incomplete or inadequate.