This paper addresses the problem of camera calibration using object silhouettes in image sequences. It is known that silhouettes encode information on camera parameters by the fact that their associated viewing cones should present a common intersection in space. In this paper, we investigate how to evaluate calibration parameters given a set of silhouettes, and how to optimize such parameters with silhouette cues only. The objective is to provide on-line tools for silhouette based modeling applications in multiple camera environments. Our contributions with respect to existing works in this field is first to establish the exact constraint that camera parameters should satisfy with respect to silhouettes, and second to derive from this constraint new practical criteria to evaluate and to optimize camera parameters. Results on both synthetic and real data illustrate the interest of the proposed framework.