A robot-arm system which is able to manipulate a moving object on a belt-conveyor at various speeds was built. This system consists of two parts. One replaces (grasp and put) a moving object based on fuzzy-interference rules with the aid of image processing. The second part is related to recognizing patterns. In this part, a method of constructing a discriminant tree is proposed, where three newly defined measures called effectiveness, importance, and applicability are introduced. The robot-arm system is able to recognize the shape and the size of moving patterns on a belt conveyor based on the discriminant tree. The whole system is controlled by only one 16-bit microcomputer, and works in real-time. It also allows human-like flexible movement. The advantages of the proposed method are reduction of processing time and availability of low level devices.