A　sensing　and　controlling　system　is　established　for　pulsed　gas　metal　arc　welding　(P-GMAW).　The　topside　weld　pool　characterizing　parameters　can　be　captured　in　real-time　and　the　back　side　weld　pool　width　can　be　predicted　from　these　parameters　by　a　neural　network　model.　A　fuzzy　controller　is　established　and　then　the　close　loop　controlling　for　weld　penetration　is　realized.　Based　on　some　topside　weld　pool　neural　networks　and　the　back　side　weld　pool　width　neural　network,　a　simulation　system　is　constructed　and　the　results　suggest　the　fuzzy　controller　can　meet　the　requirements　of　weld　shape　process　controlling.　The　variable　gap　experiments　also　validate　the　performance　of　the　fuzzy　controller;　the　maxim　controlling　error　is　0.47mm.　©　2011　IEEE.