We propose an new divide-and-conquer method for solving large-scale Traveling Salesman Problem. The main feature of the method is to construct some child-problems and child-tours combined with some local parts of a tour. A solution is optimized widely by repeating optimization of each child-tours. Child-tours can be optimized individually, so our method may be applied parallel computing. In experiments, the method Local Clustering Organization is used to optimize child-tours. Reasonably approximated solutions are obtained in useful computational times. And our method shows superioritiy to large-scale TSP in convergence time of solution.