This　paper　presents　a　learning　cerebellar　model　to　control　reaching　movements　of　a　simulated　biomimetic　manipulator.　Utilizing　the　servo　mechanism　of　the　spinal　reflex　circuitry,　the　model　embeds　a　neural　network　based　on　known　cerebellar　circuitry　in　a　simulation　of　the　mammalian　motor　control　system　to　control　a　6-muscle　2-link　planar　arm.　The　system　had　implemented　a　biologically　version　of　the　parallel　hierarchical　control　model　proposed　by　Katayama　and　Kawato　and　was　proved　to　be　able　to　learn　accurate　trajectory　control.　The　simulation　results　demonstrate　that　this　cerebellar　model　was　able　to　learn　parts　of　the　inverse　dynamics　model　not　provided　by　the　PDF+F　controller,　and　having　nicer　tracing　performance　of　desired　trajectory.