Vision systems with multiple video cameras is expected to be increasingly applied to a variety of fields such as robotics and human interface owing to their wide visual field and potential for recovering 3D structure of the scene. We have developed a simple technique for calibrating such multi-camera systems using a plane with a known 2D pattern as a reference object. Not only the intrinsic parameters, e.g. focal lengths and lens distortions, of the cameras but also the relative displacement between them are simultaneously estimated with high accuracy by simply showing the plane placed at three or more locations. Thus the algorithm yields a handy and flexible means for calibrating stereo vision systems with the arbitrary number of cameras.