There are a number of modern approaches that can be used to characterize flaws in materials. For example, one method, which has been described recently by Wormley and Thompson [1], uses a model-based approach to obtain the “best fit” size and orientation parameters based on a simple equivalent shape such as ellipsoid. Before such sizing estimates can be activated, however, it is first necessary to determine if the unknown flaw being examined is a volumetric or crack-like flaw, since the sizing algorithm will be different for each case. This classification problem, although it is conceptually simpler than the more complete problem of flaw characterization, is, nevertheless, a difficult challenge because of the large number of parameters that can influence the resulting signals. A summary of our recent work on the flaw classification problem is given below. As will be shown, we have chosen to use a combination of signal processing, modeling and artificial intelligence tools to try to pare down the complexity of the ultrasonic responses and isolate those features that are dependent only on flaw-type.