In order to reduce the amount of calculation and keep the advantage of local search strategy simultaneously, this paper proposes a kind of local search method based on density for the NSGA2(NSGA2-DLS) algorithm. Firstly, the sparse degree of each solution is evaluated with the density of each solution in the solution space. Then the non-dominated solution with the smallest sparse degree is defined as the sparse solution, and the sparse solutionig is searched around locally during every genetic process. The NSGA2-DLS algorithm adopts extreme optimization strategy and random search strategy simultaneously to improve the quality of solutions and convergence rate. The performance of the NSGA2-DLS algorithm is compared with the performance of the baseline NSGA2 algorithm and two other reported local search NSGA2 algorithms for multi-objective test problems, including ZDT and DTLZ functions. The simulation results show that the solution quality and the calculation amount of the NSGA2-DLS algorithm are much better than that of other methods.