Recently, with the increase of the size of container ships, extremely thick steel plates have been applied to hatch side coaming, upper deck and other longitudinal structural members in order to maintain the structural integrity of such ships. To prevent the brittle fracture of thick plate, it is important to detect and measure the inner defects. Ultrasonic Testing (UT) is regarded as a valid method in order to inspect the inner defect of thick plate especially in welding part. In this research, the Superposing Method of waves from point sound sources (SPM) was proposed and used for calculating the reflected wave from defects in the various situations instead of FEA. In this method, the probe and defects are treated as the assemblies of sound sources. The ultrasonic propagation is expressed as the mutual transmission from each sound source. The time history of wave amplitude on the receiving surface is calculated by superposing the wave amplitude from sound sources on the transmitting surface by considering the delay and decay of wave. The calculation speed of the simulation program by SPM is much faster than that by FEA. To confirm the accuracy of SPM, the calculated results are compared with the results of FEA and experiments. Moreover, a neural network system which is used for estimating the defect attributes, such as size, shape, and location, are developed by using the numerical simulation results by SPM as training data. In order to confirm the applicability of this system, the estimated defect characteristics are compared with the true values for several test specimens.