Multicomponent acquisitions offer the opportunity to form elastic migration images and to estimate elastic parameters of the subsurface. Compared with better constrained P-wave velocity inversions, it is more difficult to estimate the S-wave velocity due to strong nonlinearities introduced by converted S-waves. We have developed an iterative S-wave velocity inversion method guided by image registration. Given an accurate P-wave velocity and a simple initial S-wave model, we form P-P and P-S images using elastic reverse time migration. We use image registration to find the shifts between the P-S and P-P images. The S-wave velocity model could be updated iteratively by minimizing the differences between the original and the fractionally warped P-S images. A simple layered model and a modified Marmousi model are used to demonstrate the viability of the new method. In both examples, high-quality P-S images, as well as smooth S-wave velocity models, are inverted efficiently with a homogeneous S-wave initial model.