To extend the capability of commonly used graphical user interface, we have been studying a novel interface which provides various supports for a user without explicit control commands. This paper describes a method for tracking a user's hand in 3D and for recognizing the hand's gesture in real-time without using any invasive devices attached to the hand. In particular, multiple camera inputs are used to overcome the problem of occlusions, and a neural network is used for fast and robust recognition of predefined hand gestures.