A new calibration algorithm for multi-camera systems using a small 3D reference rig is proposed. Extrinsic and intrinsic parameters of the cameras are recovered only by capturing the reference rig placed at different locations. Thus the proposed method yields a simple calibration means for multi-camera systems while not using large reference rig, which should occupy the image capture volume as full as possible for accuracy. The algorithm is based on factorization of projection matrices into the camera and rig pose parameters. The obtained parameters are further refined by a non-linear optimization process. The results of the experiments with real images are presented to demonstrate the feasibility of the proposed method.