To　improve　the　rate　and　accuracy　of　weld　defect　segmentation　in　radiographic　images,　a　segmentation　method　based　on　water-flooding　principle　is　proposed.　According　to　the　principle　water　flows　downwards　to　the　low-lying-land　and　runs　over　when　it　is　full,　the　region　of　a　defect　and　its　neighbor　are　flooded　and　the　threshold　value　is　determined　in　terms　of　the　growth　of　the　flooded　area.　The　waveform　analysis　of　gray　level　profiles　is　used　to　detect　a　defect　and　get　the　seed　point.　Then,　a　flooding　process　is　carried　out　starting　from　the　seed　point,　where　a　line　by　line　search　strategy　is　employed.　Once　flooding,　the　threshold　is　determined　by　analyzing　the　relative　increment　of　the　flooded　area.　The　experimental　results　show　that　the　computation　cost　of　the　proposed　method　is　5.3%　and　25%　of　that　of　Alaknanda　method　and　Kaftandjian　method,　and　the　segmentation　error　gets　27.33%　and　53.14%.