Abstract: An algorithm was proposed for defect recognition in circumferential joints of welded rotors based on PSO-SVM model. First, the denoised ultrasonic defect echo signals were decomposed by four layers wavelet packet and node reconstruction. Then, the form factor and crest factor were extracted from the approximate portion of node reconstructed signals, on which basis, the sample feature vector was formed in combination with the ultrasound integral value and the absolute value of variance as well as the RMS in details. Finally, the particle swarm optimization (PSO) algorithm was used to optimize the penalty factor and kernel function of support vector machine (SVM), thus completing the defect recognition. Results show that the PSO-SVM model has good recognition performance in the prediction of samples. Comparing with other commonly used SVM models, the PSO-SVM model has advantages in both recognition rate and recognition time.