In　order　to　enhance　the　grinding　accuracy　of　numerical　control　(NC)　grinding　machine,　the　kinematics　of　multi-body　system,　neural　net　and　finite　element　method　were　used　to　make　geometric　error,　thermo　error　and　load　error　prediction　models.　Geometric　error　and　loading　error　could　be　reduced　by　using　precision　NC　instructions　which　were　generated　through　an　inverse　kinematics　solve　method,　and　the　thermo　error　could　be　reduced　through　an　special　mini　displacement　mechanism　which　was　controlled　by　the　thermo　error　prediction　neural　net.　Experiment　results　show　that　the　grinding　shape　error　reduced　from　0.225　μm　to　0.045　μm　after　the　application　of　the　above　sythetic　error　compensation　method.