Mutual information is a widely successfully used similarity metric in multimodality medical image registration. However, it employs the global intensity statistical parameters and lacks of spatial information. In order to combine the advantages of spatial information, a regional weighted mutual information is proposed for image registration especially for multimodal rigid registration with large rotation and translation. Experiments on 132 sets of CT and MR data (including different sequences)show that the regional weighted mutual information is robust similarity metric. It can align the multimodal data with large rotations, even if mutual information fails. Because the regional weighed mutual information can be simultaneously and parallel analyzed, the computation efficiency is not inferior to the mutual information method.