In　order　to　improve　the　performance　of　gyroscopes,　the　random　drift　error　of　a　micro　electro　mechanical　system　(MEMS)　gyro　was　analyzed　and　modeled.　The　noise　feature　of　MEMS　gyro　is　analyzed　based　on　the　AR　model.　By　introducing　a　fading　factor　of　Strong　Tracking　Filter　(STF),　the　Sage-Husa　adaptive　Kalman　filter　reduced　the　effect　of　the　error　of　model　and　noise　statistical　characteristics.　The　processed　signal　of　a　certain　type　of　gyroscope　is　filtered　by　the　new　Kalman　filter.　Through　the　test　on　a　certain　type　of　gyroscope,　the　processed　result　from　the　practical　simulation　shows　the　new　adaptive　Kalman　filter　is　not　sensitive　to　the　error　of　model　and　noise　statistical　characteristics,　the　accuracy　of　drift　signal　is　improved　greatly.　©　2014　IEEE.