When objects are recognized by using multiple cameras, recognition rates strongly depend on the camera arrangement. In this paper, we propose a new method for planning a multiple camera arrangement for accurate recognition. We use a parametric eigenspace method for the recognition framework in which objects are represented as manifolds in an eigenspace. The proposed method evaluates the adequacy of camera arrangement according to the relations between the manifolds in the eigenspace. In the experiments, we defined a function that measures relations by the distances between manifolds. The experimental results show the effectiveness of the proposed method.