A divide-and-conquer approach for the feedback arc set is presented. The divide step is performed by solving a minimum bisection problem. Two strategies are used to solve minimum bisection problem: A heuristic based on the stochastic evolution methodology, and a heuristic based on dynamic clustering. Empirical results are presented to compare our method with other approaches. An algorithm to construct test cases for the feedback arc set problem with known optimal number of feedback arcs, is also presented.