The　spectrum　of　testing　signal　for　the　spot　weld　test　is　analyzed　and　the　characteristic　vector　which　can　indicate　the　characteristics　of　the　signal　spectrum　of　ultrasonic　signals　for　spot　weld　is　obtained.　Through　using　the　vector　as　input　data,　an　artificial　neural　network　is　proposed　to　classify　the　resistance　spot　welds　in　the　different　diameter　level.　The　testing　method　proposed　in　the　paper　has　the　advantages　of　higher　recognition　ability,　higher　efficiency　and　smaller　interference　factors　compared　to　the　traditional　methods.