In order to support the ultrasonic inspection technique of defects, a method that combines the expert system with the neural network is proposed. Knowledge of selection of the NDT inspection and ultrasonic testing methods is expressed by the rule base in detail. Recognition knowledge of the detected ultrasonic wave signals is difficult to extract, so the recognition of the waveform reflected from defects is carried out using the neural network. Setting of the initial values is performed using a knowledge base. Many learning iterations of the neural network may be required, and sometimes the neural network may not converge. We aim to construct a flaw detection system based on a newly developed knowledge base. The usefulness of the system is verified by the improvement of the learning speed of this system. The results of learned data in terms of detected waveform are added to the case-base as object-oriented knowledge expressions, which may be used to support the nonprofessional user.