We suggest a hybrid expert system of memory and neural network based learning. Neural network (NN) and memory based reasoning (MBR) have common advantages over other learning strategies. NN and MBR can be directly applied to the classification and regression problem without additional transform mechanisms. They also have strength in learning the dynamic behavior of the system over a period of time. Unfortunately, they have an achilles tendon. The knowledge representation of NN is unreadable to human being and this 'black box' property restricts the application of NN to areas which needs proper explanations as well as precise predictions. On the other hand, MBR suffers from the feature-weighting problem. When MBR measures the distance between cases, some features should be treated more importantly than others. Although previous researchers provide several feature-weighting mechanisms to overcome the difficulty, those methods were mainly applicable only to the classification problem. In our hybrid system of NN and MBR, the feature weight set calculated from the trained neural network plays the core role in connecting both the learning strategies. Moreover, the explanation on prediction can be given by presenting the most similar cases from the case base. In this paper, we present the basic idea of the hybrid system. We also present an application example with a wafer yield prediction system for semiconductor manufacturing. Experimental results show that the hybrid system predicts the yield with relatively high accuracy and is capable of learning adaptively to changing behavior of the manufacturing system.