We propose a new real-world pointing interface, called Arm-Pointer, for user interaction with real objects. Pointing at objects for which a computer is to perform some operation is a fundamental, yet important, process in human-computer interaction (HCI). Arm-Pointer enables a user to point the computer to a real object directly by extending his arm towards the object. In conventional pointing methods, HCI studies have concentrated on pointing at virtual objects existing in computers. However, there are the vast number of real objects that requires user operation. Arm-Pointer enables users to point at objects in the real world to inform a computer to operate them without the user having to wear any special devices or making direct contacts with the objects. In order to identify the object the user specifies, the position and direction of the arm pointing are recognized by extracting the user’s shoulders and arms. Therefore, an easy-to-use real-world oriented interaction system is realized using the proposed method. We developed an algorithm which uses weighted voting for robust recognition. A prototype system using a stereo vision camera was developed and the real-time recognition was confirmed by experiment.