After　drastic　simplification　process,　most　of　the　existing　simplification　algorithms　based　on　edge　collapse　do　not　preserve　important　shape　features　of　the　original　model　very　well　and　lead　to　visual　distortion.　This　paper　presents　two　edge-collapse　based　algorithms　using　improved　quadric　error　metrics.　Two　important　degrees　of　a　triangle　are　given　and　embedded　into　the　original　Garland＇　s　quadric　error　metrics,　so　that　the　metrics　can　not　only　measure　distance　error　but　also　reflect　geometric　variations　of　local　surface.　The　experimental　results　show　that　the　novel　algorithms　can　preserve　quite　a　number　of　important　shape　features　and　reduce　visual　distortion　even　after　performing　drastic　level　of　simplification.