Multiple local minima exist in almost every coplanar or non-coplanar radiotherapy treatment planning problem. We used a global optimization method based on topographical information about the distribution of local minima to find all local minima and select the best as the global minimum. We uniformly select N random points from search regions and construct a topographical graph, from which M (M ≪ N) starting points are selected to launch a local search. Because every seed point was at or near the local minimum, the solutions found by the local search could be used as the final optimization results. We verified this algorithm by applying it to three different clinical cases and comparing the results with those obtained by a local optimization method (sequential quadratic programming). The results show that this algorithm is feasible and efficient.