This paper solves the multobjective optimization for the VRPbased NSGA II. First of all, we analyses and generalizes three object functions and three constraints, and realize the mathematical modeling for VRP. Then we code in MATLAB, and choose the data in C101 of Benchmark Problems as the test data. Furthermore, we improve the determining of initail population and the crossover operator of NSGA II, against the shortcomings of design in NSGA II. By comparing the two algorithms, we verify the improved algorithm is effective in avoiding premature convergence, raising efficiency and stability.