The　sharp　features(1),　including　edges,　corners　and　boundaries,　usually　express　the　most　important　geometric　information　for　triangular　meshes.　The　purpose　of　this　investigation　is　to　reconstruct　sharp　edges　from　blended　or　chamfered　features　for　mesh　processing　with　high　accuracy.　The　proposed　approach　involves　two　processes:　sharpness-based　region-recognizing,　which　identifies　the　feature　vertices　based　on　the　vertex　sharpness,　and　sharp-features　reconstructing.　In　the　process　of　region-recognizing,　the　angle　variations　of　normal　vectors　are　introduced　to　indicate　the　vertex　sharpness.　Then,　the　feature　regions　comprised　of　＂sharp＂　vertices　and　facets　could　be　identified.　Furthermore,　during　the　process　of　sharp　feature　reconstructing,　the　coordinates　of　concerned　vertices　are　adjusted　gradually　using　an　iterative　filtering　algorithm　depending　on　sharpness,　which　updates　the　feature　regions　from　the　inside-outside.　Finally,　the　experimental　results　validate　the　effectiveness　and　robustness　of　the　proposed　method　in　sharpening　meshes.