A　method　based　on　cerebellar　model　with　feedback-error-learning　was　proposed　to　control　a　two-wheeled　self-balancing　robot.　The　forward　controller　which　is　an　on-line　neural　network　adaptive　controller　is　corresponding　with　cerebellar　and　expresses　the　function　of　cerebellum　in　balancing　control　problem.　The　robot　is　a　kind　of　MIMO　system,　so　the　cerebellar　cortex　part　has　two　nerve　cell　output.　From　the　simulation,　we　can　see　that　the　robot　can　be　balanced　in　fixed　position　well　by　this　method,　and　can　improve　the　feedback　control　to　some　extent.　It　is　a　kind　of　bionic　control　method　which　imitates　the　cerebellum　part　of　brain.　©2010　IEEE.