After　study　of　existing　point　cloud　denoising　algorithm　carefully,　the　normal　vector　and　local　square　error　in　normal　direction　were　computed　for　point　sets.　A　self　adaptive　optimal　denoising　neighbor　which　was　related　to　the　surface　feature　was　achieved　by　using　a　self　adaptive　angle　threshold　trimming　function.　Normal　space　denoising　and　location　space　denoising　were　carried　out　by　using　improved　trilateral　filter.　A　series　of　experiments　work　well,　contrasting　the　new　denoising　algorithm　herein　and　other　ways,　more　features　have　preserved　and　smoothing　surface　have　been　achieved.