The objective of this work is to develop an advanced automatic ultrasonic inspection system via adaptive learning network signal processing techniques. This system will provide the type, location, and size of defects in metal more quickly and to smaller defect size than current imaging systems, without the need for operator interpretation of the results. An ultrasonic imaging array constructed for this project has been used to record data from artificial defects in carbon steel test blocks. Software has been written to automatically determine the orientation and size of cracks from these digitized waveforms. Detection of these cracks has been unambiguous down to 1/6 wavelength or 0.25 mm. Sizing for depth is accurate to 12% down to 1/3 wavelength. Further research will extend these results to other defect types and to smaller defects. The significance of this work is that it will demonstrate the feasibility of a totally automatic detection, classification, and sizing system which will work with hardware ordinarily used for imaging. This system will provide a numerical estimate of the defect parameters rather than an image requiring operator interpretation, and it will do so at defect dimensions smaller than the limits set by the resolution of imaging systems.