Abstract: 3D pre-stack depth migration needs an accurate velocity model to image subsurface structures. As an inverse problem, migration velocity analysis tries to infer the velocity model from the picked residual curvatures on common image-point gathers (CIGs). Because of noisy data, the linearization of nonlinear problems, and inadequate parameterization of the model, migration velocity analysis is thought to be ill-posed and ill-conditioned. Reasonable solutions can be obtained only by adequately defined regularization. Firstly, the basic aspects of the Bayes’ theorem and the classical “information-based” maximum-entropy method were reviewed. Then, the classical maximum-entropy method was extended by an incorporating model derivative into a maximum-entropy computation. This has the effect of intra-region smoothing including the preservation of sharp boundaries. Finally, a numerical example proves the effectiveness of the proposed algorithm.