Predicting annual runoff is very important for guiding the management of water resources. Annual runoff is a result caused by many factors, and the relations of annual runoff and the predicting factors are complex and nonlinear. Nowadays the annual runoff prediction models mostly express the relations by explicit functions, the expressions of the functions will be changed according to the studying areas, and it is very difficult to resolve the functions. Predicting results in practice are generally obtained by comparing the predicting factors values of the current annual runoff with the sample series of the history annual runoff in the same area, which is proved to be very feasible and effective by a lot of practices. For this reason, a new model for predicting annual runoff, named SP model, is presented based on Shepard interpolation technique. The case study shows that SP model is convenient and effective, and that SP model can be applied to mid and long term prediction of many runoff processes