Abstract: The non-dominated sorting in genetic algorithms (NSGA) has some deficiencies such as the poor local search and premature convergence. An improved algorithm based on the advantage of simulated annealing is presented to overcome these shortcomings. The local search operator of simulated annealing for multiobjective optimization and the jump criteria are taken part into the new algorithm. The local search should be carried out by simulated annealing in the vicinity of the 1st and 2nd rank of non-dominant solutions. This approach can improve operational efficiency and make up for the deficiencies of NSGA. The simulation results show the effectiveness of the algorithm.