We describe DIRECT-BP, a modi cation to the DIRECT algorithm that attempts to perform a more balanced search about the best point found in each iteration. DIRECT ensures the division of one of the largest boxes in each iteration, which asymptotically guarantees convergence to a global optimum. We demonstrate that DIRECT can fail to e ectively perform local search about the best point found even though it globally optimizes. DIRECT-BP modi es the DIRECT algorithm to ensure more e ective local convergence. DIRECT-BP forces a subset of the boxes neighboring the best box to divide, and it provides a mechanism for the local search to move from one box to another. This allows continuous progress towards a local optimum without relying on the global aspects of DIRECT.