In order to study the output loads of TN (total nitrogen) and TP (total phosphorus) from agricultural non-point sources in Cao'e River basin in Shaoxing, an improved export coefficient model for TN and TP was firstly established. According to the theory of black box, the TN and TP from rainfall and agricultural pollution sources were respectively integrated to the total amount of pollutant in this model, and the output intensity of TN and TP in the transportation process was taken into account for the pollutant producing coefficient for the basin, which reflected the effects on the output of TN and TP due to the various losses during the process of rainfall runoff and pollutant transport. After that, the total output loads of TN and TP from agricultural non-point sources during 2005-2010 were estimated using this improved export coefficient model, in which the basic data were obtained from Shaoxing Statistical Yearbook. When the values predicated by the model were compared with those measured by the experiments, the authors found that the total output loads of TN and TP from agricultural non-point sources during 2005-2010 were greatly influenced by rainfall. The total output loads of TN and TP had a positive exponential relationship with the annual rainfall intensity, and therefore a new improved export coefficient model for TN and TP based on rainfall intensity was also established. Finally, this new model was applied to estimate the output loads of TN and TP from various agricultural pollution sources in Cao'e River basin. The estimation results showed that the total output loads of TN were annually far greater than that of TP from agricultural non-point sources during 2005-2010, and the maximum was up to 20.67 times, but with the influence of annual rainfall intensity, they presented uneven temporal distribution during 2005-2010, and the annual total output loads of TN ranged from 5456.60 to 12268.38 t and those of TP ranged from 393.19 to 820.65 t. In those years, when the rainfall intensity was small, the total output loads of TN and TP were both relatively low, and the contributions of rainfall to output loads of TN were up to 54.75%-69.67%, indicating that rainfall had become a key pollution source for TN output in Cao'e River basin. In addition, the results also showed that the order for the contributions of various agricultural pollution sources to output loads of TN from high to low was as follows: rural residents, livestock and poultry, and agricultural land. As a result of shortage of environmental infrastructure in rural area, deficiency of life sewage collection pipe, discharging the polluted water into river directly, and letting off waste randomly, rural life became primary agricultural pollution source to the output of TN in Cao'e River basin. However, the contributions of various agricultural pollution sources to output loads of TP decreased in the order of livestock and poultry, rural residents and agricultural land, and the contributions to output loads of TP from livestock and poultry (44.75%-55.74%) were significantly greater than that from rural residents in Cao'e River basin, showing that the problem of fecal pollution should be paid enough attention with the rapid development of livestock and poultry breeding industry. In comparison, the contributions to output loads of TN and TP from agricultural land were smaller, but the output loads of TN and TP from cultivated land were annually greater than that from garden plot and forest land. Therefore, it was necessary to further control the excessive fertilization in agricultural land to reduce the loss of nitrogen and phosphorus. The improved export coefficient model for TN and TP in this paper had certain applicability in Zhejiang area, however, in order to further improve the accuracy of the model, it was still necessary to strengthen the monitoring of regional rainfall runoff and the study on the uncertainty of model parameters in the future.