A multi-resolution multi-mode medical image registration method is proposed to improve the speed and accuracy of registration. The images were decomposed by wavelet function, and expanded Powell optimization algorithm was used to find the maximal mutual information (MMI) between images. The registration starts from the lowest-resolution image level 0 of the image pyramid, the deformation parameters of image level 0 registration are available as the initial parameters of image level 1, repeating the process until the completion of the highest-resolution image registration. The image registration results of CT/MRI and CT/PET show that wavelet-based multi-mode registration method has advantages such as avoiding local maxima, fast convergence and high accuracy.