Abstract: A novel genetic algorithm for TSP was proposed for the defections of traditional GA which are converging very slowly and easy to premature.It merged two heuristics as crossover and mutation respectively which can improve the local search ability of GA, and then imported the exterior best-individual set to increase the diversity of the population.And further tests for the proposed Algorithm are applied.Based on these, the algorithm is parallelized by the recursively divide-and-conquer method.The simulation results indicated that it can get high quality solution and cost litter running time.