In mobile robot application, many map building algorithms using ultra-sonic sensors were introduced. In this paper, a new feature extraction method using neural networks is proposed. In indoor environments, the ultrasonic sensor system has some uncertainties in its data. To overcome this situation, a neural network approach is used. The up-date region is changeable according to distance data of sensors for the symmetry of the location of sensors. Target differentiation method is implemented using fewer ultrasonic sensors. The reflection wave data patterns are learned using neural networks. Therefore, with neural net-works the targets are classified to plane, corner and edge roughly that frequently occur in indoor environment. We construct our own robot system for the experiments and some experiments are carried out to show its performance.