Abstract: The estimation of depth velocity model is a great challenge in seismic imaging, obtaining an accurate depth velocity model and reducing the depth of imaging project turnaround time are of critical importance. The velocity modeling of conventional tomographic inversion is equivalent to a linear inversion, and need to be re-picked up, which leads to very time consuming and low efficiency. A velocity modeling method based on nonlinear tomography is proposed. The RMO of the common imaging point gathers is converted to the pre stack domain as the kinematic invariant, and the model updating is carried out by the iteration of the tomographic inversion. A multiple linear inversion is used to approximate a nonlinear physical process, which avoids the repetitive work and greatly improves the efficiency of project. The validity of method is proved by an example.