A clustering algorithm which arranges paths after grouping was proposed to solve Split Delivery Vehicle Routing Problem (SDVRP). Considering the balance of vehicle load and characteristics of feasible solution, first, the customers which load were greater or equal to the vehicle load limit were arranged in advance. Then combined with the distance between customers and load, a split threshold was set to limit the load of vehicle to a certain range. According to the nearest principle, all customers were clustered and grouped. If the customer load in a group does not reach the minimum load of vehicle and is beyond the limit load when new customers are added in, the new customers were split and adjusted. Finally, while all the customers were divided into groups, the customers paths for each group was arranged by Ant Colony Optimization (ACO) algorithm. The experimental results show that the proposed algorithm has higher stability, and better results in SDVRP.