A　kind　of　defect　recognition　algorithm　based　on　BP　neural　network　is　used　to　extract　characteristic　quantities　from　the　samples　obtained　at　different　experimental　conditions,　and　distinguish　the　different　sizes　and　positions　of　radial　cracks　in　steel　rods.　Firstly,　the　axisymmetric　longitudinal　guided　waves　mode　in　235　kHz　are　excited　to　detect　the　radial　cracks　in　a　steel　rod.　Experimental　results　show　that　obtained　guided　waves　contain　quite　single　L(0,2)　mode　in　235　kHz,　it　avoids　the　problem　of　weaker　detection　capability　using　L(0,1)　mode　to　detect　small　size　defects,　and　reduces　the　difficulty　to　distinguish　the　defect　echo　because　more　modes　involved　when　axisymmetric　longitudinal　high　order　modes　are　used　to　detect　the　steel　rod.　Secondly,　the　algorithm　is　used　to　recognize　the　radial　cracks　in　a　steel　rod.　Results　show　that　the　proposed　defect　recognition　algorithm　can　identify　different　depths　and　positions　of　cracks　well,　and　the　correct　rate　of　recognition　has　stabled　at　87%　in　existing　experimental　samples.