Abstract: Abstract: To solve problems of lack of water resources for agriculture in arid areas, it is necessary to utilize and compare the advantage of water using in different areas that are linked together through the commodity trade. The active regulation of commodity trade and water consumption in production is capable to build rational virtual water use embodied in commodity trade. It will ease the conflict between water supply and demand. This requires the implement of integrated management about real water and virtual water to achieve regional equilibrium of virtual water utilization. Considering the water utilization should fit the scale of water resources endowment and economy, the equilibrium level based on Gini coefficient is used to evaluate the regional equilibrium of virtual water flow against water resources endowment and Gross Domestic Product. To improve the regional equilibrium, a comprehensive regulation model about agricultural water consumption was setup. The decision variables belonged to the aspects of water use efficiency, plant area of grain and yield level per unit area. Through the adjustment of the decision variable, the relevant equilibrium level can be calculated using a sub-model. In the sub-model, based on the equilibrium analysis of grain supply-demand, the grain flow direction and quantity were estimated using the optimization model for minimizing the total transportation costs. Combined with the agricultural water consumption per unit mass evaluation on every district, the amounts of virtual water flow among districts, and the total amounts of virtual water inflow and outflow of every district were setup. Then the equilibrium level about water resources endowment and Gross Domestic Product were calculated and became the output of the sub-model. To minimize the value of equilibrium level, the particle swarm optimization was used to search the non-inferior solutions about the combination of various decision variables. Then the final solution was selected in the non-inferior solutions. To test the performance of the method, a case study was made in Gansu Province in semi-arid Western China. The grain supply and demand, agricultural water consumption and water resources endowment of every county and city in Gausu province were estimated. They were the inputs of the comprehensive regulation model. The decision variables included planted areas, irrigation water amounts per unit area, and the yield per unit areas of wheat and corn in every district. The planted area can increase or reduce within a certain range, the irrigation water amounts per unit area and the yield per unit areas can only have a certain amount of reduction and growth respectively. The objective was that the agricultural water consumption in every district after regulation should not increase. The transportation of corn and wheat were railway and highway, and the specific means of transportation between every two districts was determined by the lowest cost per unit mass. By using the comprehensive regulation model, the non-inferior solutions gave the various combinations about decision variables. The grain areas, grain productions, water consumptions and virtual water flows in different solutions were quite different, but the equilibrium increased significantly comparing to that before regulation. In these solutions, the overall pattern about the virtual water flow in Gansu did not change a lot, and the virtual water import and export districts kept unchanged. The significant change emerged in the grain production and virtual water flows in some districts. The results showed that the agricultural water comprehensive regulation was capable to improve the equilibrium of virtual water matching the water cap, and can be a useful supplement to existing integrated water resources allocation methods. Further studies about the comprehensive regulation need to focus on the introduction about more realistic constraints, finer regulation mechanism and more accurate flow of grain trade.