In this article, an improved method to segment the renal cortex and medulla and to eliminate the influence of kidney motion produced by respiration is proposed. On the basis of an adaptive threshold estimated from the mean gray levels of the grown regions, a region-growing algorithm is presented to produce a three-dimensional (3-D) kidney contour and to segment the renal structures. Moreover, a shell mask of kidney margin is proposed to realize a coarse matching so as to eliminate the image translation, which makes the processing simple and direct in spatial processing without any image transform computations, and a correlation computation can be implemented with great efficiency. Then, a refined matching with a mask of cortex is completed through a 3-D correlation algorithm to ensure the accurate registration of the images in different phases. Comparing with the global mask including the whole kidney, both the shell mask and the cortex mask significantly contribute to decreasing the matching errors for images with nonuniform intensity signals, which much improves the registration quality of renal MR images. In this way, the effect of the respiration motions can be eliminated so that the intensity measurement in different phases becomes accurate within the respective structures.