There has been an interest to apply neural network as inference engines of the medical consultation system, because the neural expert system is expected to solve some obstacles of the knowledge acquisition. However, some problems remain to be examined in development of neural expert systems. One problem concerns how the data structure influence on the process of the knowledge acquisition and what kind of data structure is desirable in the learning process. Another one concerns that the knowledge embodied in neural network is difficult for medical experts to understand the meaning of the connection weights. The purpose of this study is to elucidate these problems concerning the knowledge acquisition by the neural network.