In　the　field　of　practical　engineering,　the　moving　least-squares　(MLS)　and　moving　total　least-squares　(MTLS)　methods　are　widely　used　to　approximate　the　measurement　data.　However,　the　fitting　accuracy　and　robustness　are　affected　inevitably　by　random　errors　and　outliers,　which　results　in　distorted　estimation.　To　solve　this　problem,　we　present　an　improved　MTLS　(IMTLS)　method,　in　which　a　newly　constructed　parameter　is　introduced　to　characterize　the　geometric　feature　of　abnormal　points　in　each　influence　domain.　The　total　least-squares　(TLS)　method　is　used　to　generate　the　fitting　points　at　first.　The　corresponding　variance　of　each　node　is　then　calculated,　and　the　connection　between　abnormal　degree　and　random　errors　is　established　to　trim　the　nodes　with　a　certain　abnormal　degree.　The　remaining　nodes　are　used　to　find　the　local　coefficients.　This　proposed　algorithm　can　avoid　the　negative　effects　of　outliers　without　human　intervention　and　overelimination.　The　final　results　of　the　measurement　experiments　and　numerical　simulations　demonstrate　that　the　improved　algorithm　has　better　robustness　and　accuracy　than　the　MLS　and　MTLS　method.