Various types of defect may be formed in girth welds of long-distance pipeline in the process of welding. They are hidden dangers to pipeline transportation safety. Currently, ultrasonic phased array instrument is commonly adopted for quick automatic positioning and quantitative analysis of flaws in the girth weld after welding. But as for qualitative analysis – flaw classification, traditional manual identification method is still used. By traditional method, human-made error is easily introduced and classification result is depended on the detection experiences of the inspecting person. To overcome these deficiencies, a new method combined second generation wavelet transform (SGWT) with Radial Basis Function neural network (RBFN) is proposed in this paper, realizing automatic flaw classification and reducing human factors impaction. SGWT is ideally matched local characteristics of the flaw signal, improving both the computational speed and flaw classification efficiency. Then, based on the “energy-status” feature extraction method and the above SGWT analysis, feature eigenvectors of the flaw signals are extracted, training the following RBFN. And then when the feature of any flaw is extracted, it can be recognized by the network. The output of the network is the type of the input flaw signal, realizing automatic flaw classification. Finally, an ultrasonic phased array inspection system is described. The system is integrated with automatic flaw detection and classification. Experiments are tested on a long-distance pipeline girth weld block with artificial defects in it. The results validate that the proposed method is efficient, which is helpful to increasing inspection speed and reliability of flaw inspection for long-distance pipeline girth welds.