Taking the Guangyinsi watershed as an example , soil erodibility K values were calculated using EPIC model at large sampling densities , and the dist ribution map of K values was made using inverse distance weighting spatial interpolation. Classical statistical analyses show significant spatial variability of soil erodibility K values at a scale of small watershed , the variance coefficient of which is 39.94%. Therefore , the spatial variability should be taken into account in soil erosion monitoring at a scale of small watershed. The relative error of K values is less than 25% when half of the total samples , 4 samples persquare kilometer , are taken.