In　order　to　solve　the　selection　of　effective　singular　values,　first　the　characteristic　of　the　singular　values　of　both　ideal　signals　and　noise　signals　was　analyzed,　a　determination　method　of　least　squares　error　was　proposed.　Then　the　impact　on　signal　de-noising　by　analyzing　different　number　of　singular　values,　indicated　that　the　number　of　effective　singular　values　were　accurate　and　reliable　by　using　the　determination　method　of　least　squares　error.　Finally,　the　method　had　been　successfully　applied　to　the　sensor　signal,　obtained　the　Feature　curve　with　smaller　error　and　more　accurate　results　in　the　determination　of　glucose　concentration.　The　results　show　that:　the　extracted　signals　have　higher　SNR　and　better　integrity.