Vertex　normal　in　triangular　meshes　is　an　essential　surface　attribute　for　point-based　smoothing,　mesh　simplification　and　feature　detection　in　fields　as　machine　vision,　virtual　reality　and　reverse　engineering.　In　order　to　investigate　how　weights　of　the　algorithms　for　vertex　normal　estimation　influence　the　vertex　normal　accuracy,　theoretically　analyses　and　experiments　for　algorithms　by　comparison　are　used　in　this　paper.　Firstly,　advantages　and　disadvantages　of　five　representational　algorithms　are　discovered　by　comparing　the　characteristics　of　their　weighting　factors.　Next,　experimental　approaches　on　sphere　surface　are　presented　to　construct　a　series　of　special　triangular　meshes　for　testing.　Experimental　results　show　that　accuracy　of　normal　estimated　is　influenced　significantly　by　the　corresponding　triangles＇　quality,　and　rational　fashion　of　weights　with　incident　angle　multiplication　and　area　reciprocal　improve　the　vertex　normal　accuracy　effectively.