The study assessed soil erosion features in hilly agricultural watersheds in central Sichuan Province, China. Relationships among area percentages of land covers, soil erosion modulus, soil loss rate, and area percentage of erosion in the watersheds were examined by multivariate regression analysis. The first two regression models were constructed with area percentages of land covers as independent variables and soil erosion modulus or soil loss rate as dependent variables. The third one was built with soil erosion modulus as dependent variable and area percentage of the four erosion intensity classes as independent variables. Results showed that for flat and slope fields, forests, shrubs, abandoned fields, and other land covers, percentage of soil loss were -6%, 89%, -4%, 11%, 12%, and -3%, and area percentage of erosion were 1%, 72%, 0%, 18%, 9%, and 0%. The slope fields were the main source of soil loss. In contrast, flat crop fields, forests and other land covers could conserve soil from eroding. Changing land covers and reducing area of slope fields to 7% and of abandoned fields to 0%, were proposed as a management strategy for soil conservation in the watersheds. As completion of the land cover conversion, percentages of the land covers should be 35%, 7%, 5%, 45%, 0%, and 8% for flat and slope fields, forests, shrubs, abandoned fields, and other land covers. The study suggested that statistical models could be successfully used for processing inventory data for make decisions in soil conservation.