Most of the improvements on the ant colony algorithm are based on the algorithm itself or combined with other algorithms, which underutilizes information of the problems to be solved, so the effect is not ideal. An object-oriented multi-role ant colony optimization(OMACO) algorithm is proposed to deal with the characteristics of traveling salesman problem(TSP) on the basis of ??-means, which divides the cities into different categories with full use of the TSP spatial information and divides the colony into different roles. Each role solves the problem independently according to their respective strategies, which enhances the algorithm’s search capability and improves the quality of solution. After each iteration, only the optimal solution of each role is applied to update the pheromone, which prevents premature convergence. The elitist strategy combined with local optima jumping is used again to the stagnation of the algorithm. Many experiments of 50 classic TSP instances show that the OMACO algorithm can obtain the known optimal solution in fewer iterations. For large-scale TSP instances, the proposed algorithm is also far better than the comparison algorithms.