Agricultural Non-point Sources Pollution is one of a leading cause of water quality problems in southeast of China. It is an effective approach for ANSP control that modeling nitrogen and phosphorus transfer in a basin system by using mathematic models based on geographic processes. In this paper, the multi-scale integrating distributed modeling frame was designed based on different spatial discretization units by using the (Soil and Water Assessment Tools) model and field-scale APEX (Agricultural Policy/Environmental eXtender) model. A large number basic geographic data were collected and parameterization in Lianshui basin, Xingguo County, Jiangxi province. The mutil-scale distributed computer simulation for water yield, sediment yield of 1991 to 2005 and the transport processes for nitrogen and phosphors of Oct, 2002 to Oct, 2004 were carried out in the study area of Lianshui basin. The average simulation accuracy, coefficient of efficiency (ENS) and deterministic coefficient (R2) were 89.4%, 0.88 and 0.89 for annual water yield and 75.1%, 0.70 and 0.76 for annual sediment yield, respectively. The simulation results indicate that the integrating modeling has an acceptable performance in prediction of water yield and sediment loading in Lianshui basin. The predicted monthly nitrogen and phosphorus loading have larger amount difference than measured although both the predicted and measured have almost accordant trends. The reasons that caused bad accuracy for nutrient simulation should be further studied. The method of integration for multi-scale models can take the advantages effectively for each model which can be applied in quantitative study on the geographical processes in different levels.