Soil erodibility K-factor is an indispensable parameter in soil erosion prediction models, such as USLE and RUSLE. Immeasurable error might result from direct indiscriminate use of these empirical models in estimating K-factor. Based on observed K factors of seven typical soils in subtropical China, uncertainties of five K-factor prediction models (i.e., nomogragh model, modified nomogragh model, EPIC model, Geometric mean particle model, and Torri model) are evaluated by means of statistics such as mean absolute error(MAE), mean relative error(MRE), root mean squared error(RMSE) and accuracy factor (A_f). Results show that the five models could be lined in the order of Torri modelMRE for the Torri model is only 0.291, it is still high in uncertainty. However, an optimized Torri model can minimize the uncertainty. Its linear regression coefficient between observed K factors and predicted K factors is b =1.028 (R²＝0.921，^p<0.01), with an MRE being 0.120. It is therefore applicable to prediction of soil erodibility K-factor of certain soils in subtropical China.