A　new　on-line　adaptive　control　scheme　based　on　feedback-error-learning　is　proposed　and　applied　to　inverted　pendulum　balancing.　The　proposed　adaptive　controller　for　balancing　consists　of　a　conventional　feedback　controller　(CFC)　and　a　neural　network　feedforward　controller　(NNFC).　In　the　NNFC,　the　feedback　error　signal　is　employed　as　input　stimulator,　instead　of　the　usual　reference　signal.　An　on-line　back-propagation　(BP)　algorithm　with　the　self-adaptive　learning　rate　is　developed　and　employed　in　the　NNFC　to　realize　the　combination　of　learning　and　controlling.　Computer　simulations　on　inverted　pendulum　balancing　task　demonstrate　that　the　proposed　adaptive　controller　could　effectively　reduce　precision　requirements　of　the　CFC　parameters,　and　guarantees　good　balance　performance　and　acceptable　robust　performance.　©　2006　Elsevier　B.V.　All　rights　reserved.