Abstract: Combining mutual information of an image with its gradient information, a new bilateral image registration method is developed to use the image's spatial information conventional registration methods neglect. Taking an image A as a reference, the gradient similarities are calculated between the corresponding points of image A and B to be registered after a Gaussian operator is introduced to reduce the effect of noises, and the normalized mutual information multiplied by the gradient similarity is named "forward measure". Conversely, using the image B as a reference, the gradient similarity is thus calculated, thereby "backward measure" being determined. Then, the normalized mutual information combined with bilateral gradient similarity is obtained. Experimental results show that the new method is more robust and accurate than the conventional ones.