The　present　paper　presents　a　method　for　weld　location　extraction　in　radiographic　images.　Images　are　processed　line　by　line　by　applying　fuzzy　reasoning　based　on　local　pixel　characteristics.　For　each　pixel,　values　of　spatial　contrast　and　spatial　variance　are　computed　for　evaluating　the　edge　fuzzy　membership　value.　The　method　proposed　uses　the　machine　learning　approach　for　knowledge　acquisition,　which　automatically　generates　fuzzy　rules　by　learning　from　examples.　Using　this　method,　all　welds　are　successfully　extracted　from　101　radiographic　images　taken　from　aluminium　alloy　welding　parts.