It is sensitive to the initial population while the genetic algorithm (GA) is used to solve the traveling salesman problem (TSP). To overcome this problem, the neighbour field method is presented to create initial population. In this method the next city is not the nearest asyetunvisited location but randomly selected from the unvisited cities in neighbour field. Neighbour filed method can extract the local optimal information of adjacent cities, and the constructed population has the diversity character. Comparing to the random initial method, the mean value obtained by the neighbour field method in four standard test instances of TSPLIB improved by 46.3%. The simulation results show the effectiveness of the neighbour field method for creating the initial population.