Abstract: Abstract: Eutrophication caused by the enrichment of nutrients from diffusing sources is degrading surface water quality throughout the world, while an increased loss of agricultural nutrients is a growing concern for water quality in drinking water areas of Beijing. Best management practices (BMPs) have been proven to actually reduce nitrogen (N), phosphorus (P) and sulfur (S) pollutant loads from agricultural areas. However, in a watershed with multiple farms and multiple BMPs feasible for implementation, it becomes a daunting task to choose the right combination of BMPs that provides maximum pollution reduction with least implementation costs. Several studies have shown that best BMPs are effective in controlling water pollution. However, those issues affecting water quality need water management plans that take into consideration BMPs selection, placement and affordability. In this study, a framework of "Risk assessment - Planning and zoning - differentiated management" was developed, and it included three tools: 1) A new "risk assessment" tool was introduced for potential loads estimation of N, P and S pollution in BeiZhai small watershed by analyzing social economic data, land use, soil type, water and soil conservation practices and agricultural management measures under current conditions, and then the critical source area was identified according to the pollution loads based on GIS technology; 2) A multi-criteria index ranking system for the BMPs was devised. First, each individual second-level index was assigned a numeric value that was based on site characteristics and information on LIDBMPs. The quantified indices were normalized and then integrated to obtain the score for each first-level index. The final evaluation score of each BMP was then calculated based on the scores for the first-level indices. Finally, the appropriate BMP types for a specific installation site were determined according to the rank of the final evaluation scores, and furthermore the final scores could be served as a first screen and selective reference for the BMP placement and optimization in next step; 3) Three typical areas with different spatial scales were extracted from the BeiZhai small watershed, and a Non-dominated Sorting Genetic Algorithm (NSGA-II) was selected as an optimization engine to evaluate the optimal fitness of each BMP combination based on the initial pollutant loadings, targets of pollutant reduction and the costs of BMPs implemented at different spatial scales. The results indicated that: 1) Potential risk evaluation of non-point source pollution in the study area proved that the potential of non-point source pollution was closely related with land use patterns affected by human activities; 2) Comprehensive index values for different measures in a descending order were constructed wetlands, infiltration basin, green roof, wet detention pond and porous pavement, and these could be used as the main practices for the BMP planning; 3) The pollutants loads were reduced by 45% while the total cost (TC) was 574 560 yuan for watershed scale, the pollutants loads were reduced by 46% while the TC reached to 374 660 yuan for the community scale, the total N and P load was reduced by 65% while the TC reached to 1 518 yuan for the farm scale. Therefore, this framework can be served as a decision-making support for non-point source pollution control in the upper watershed of Huairou Reservoir.