The　weak　bonding　defect,　which　is　a　special　defect　occurring　in　solid-phase　welded　joints,　cannot　be　detected　effectively　with　normal　ultrasonic　testing　methods　used　in　industry.　The　signal　reflected　by　such　a　defect　is　so　small　that　it　is　submerged　by　the　noise.　This　weak　bonding　defect　is,　however,　a　very　dangerous　defect　and　usually　sampling　destructive　testing　methods　are　adopted　in　the　factories　to　detect　it.　In　this　paper,　numerical　signal　analysis　methods　are　used　in　the　time-frequency　domain　to　detect　the　weak　bonding　defect　as　well　as　other　defects　in　friction　welded　joints.　The　dyadic　wavelet　transform　method　and　singularity　analysis　method　are　used　to　separate　the　weak　reflected　signal　from　the　defect　from　the　noise.　Weld　quality,　weak　bonding　defects　and　lack-of-bonding　defects　are　detected　and　evaluated,　based　on　characteristics　analysis.　The　results　show　that　the　weak　bonding　defect　can　be　detected　effectively　using　a　numerical　signal　analysis　method,　and　the　lack-of-bonding　defect　in　the　presence　of　various　surface　conditions,　can　also　be　recognised.　Moreover,　more　than　one　defect　in　the　same　welding　interface　can　be　detected,　based　on　the　characteristics　obtained　by　using　a　fuzzy　logic　analysis　method.