Flaw classification is one of the essential issues in quantitative ultrasonic nondestructive evaluation of weldments. Ultrasonic flaw classification can be divided into three approaches [1]; 1) conventional approaches which use heuristic experience-based echo-dynamic pattern identification techniques, 2) model-based approaches which use model-based strong features in ultrasonic flaw signals, and 3) ultrasonic pattern recognition approaches which use features and decision making algorithms and adopt various signal processing techniques and artificial intelligent tools. Among these approaches, ultrasonic pattern recognition approaches which are considered as the most promising tool have been investigated extensively in the ultrasonic nondestructive evaluation (NDE) community [2–6].