Nonpoint source load is one of the main sources of surface water pollution. The representative annual load is the base of water pollution control. So it is very important to develop the load estimation methods of nonpoint source pollution under the limited data conditions since there are no long series of measurements on nonpoint source pollution in China. In order to meet the needs of water pollution control planning and water resources protection, on the basis of theory analysis and measured event data of nonpoint source pollution from four watersheds in the middle and the southern Shaanxi Province, with areas from 15 to 3092 km², the linear relationships between the unit load of TP, TN and unit sediment are established. Their correlation coefficients are larger than 0.9. Two examples of load prediction of nonpoint source nutrients are carried out to show the specific application procedure. One of them is the prediction of TP and TN for single extreme flood, wet year and long-term average year for the Heihe River valley with an area of 1481 km²; and another the prediction of TP and TN for different representative years for the Baihe River section (56310 km²) of the Hanjiang River, the largest branch of the Changjiang(Yangtze) River. The prediction results are compared with those of correlation method of water quality and quantity, and there are no great differences. It is concluded that this method can be applied to estimate nonpoint source nutrient loads for single event and representative years with different frequencies. Since nonpoint source pollution depends on the process of rainfall-runoff-erosion, land-use types and many other factors, the obtained relationships between nutrients and sediment will be more typical if measured data consists of high, middle and small floods. In addition, the established relationship can only be applied to the same watershed or similar watersheds to reduce prediction error of nutrient loads.