A　cerebellar　model　for　feedback　error　learning　based　on　the　Kalman　Estimator　(KECFEL)　was　developed　for　accurate　and　stable　human　body　balance　control.　The　model　uses　a　neural　network　feedforward　controller　(NNFC)　which　learns　from　the　output　of　an　estimator　and　conventional　feedback　controller　to　provide　online　learning　to　update　the　weights　and　network　training.　A　proportional　derivative　controller　is　used　as　a　feedback　controller　to　ensure　overall　stability　of　the　control　system.　Simulation　and　experimental　results　with　an　inverted　pendulum　show　the　effectiveness　of　this　model　for　human　balance　control.