To solve large-scale constraint satisfaction problems, CSPs, ant colony optimization, ACO, based meta-heuristics has been effective. However, the naive ACO based method is sometimes inefficient because the method has only single pheromone trails. In this paper, we propose an ant colony optimization based meta-heuristics with multi pheromone trails in which artificial ants construct a candidate assignment by referring several pheromone trail graphs to solve CSP instances. We also implement the proposed model to some ACO based methods, demonstrating how our method is effective for solving graph coloring problems that is one of typical examples of CSPs.