We developed and analyzed a method for automatic velocity picking in the semblance domain as a nonlinear optimization problem that is computationally fast, robust, and a simple model for testing. The obtained results can be considered as an initial model for other data-driven methods.Seismic velocity analysis can be considered the major aim for application in data imaging and post-imaging processes. It falls into several classes of mathematical and computational problems, such as manual or automatic, stack or migration, and nonlinear local or global optimization. In all cases the process needs assistance in terms of a priori information and input-output constraints, that can be geological (from well logs), geometrical, and physical parameters. In addition, all geophysics problems are to be considered three dimensional spatially, as two-dimensional imaging suffers from structural side effects.In the conventional method, the steps of velocity analysis for each common-mid-point are as follows: (1) normal-moveout stack velocities are estimated by means of semblance summation along hyperbolic time trajectories producing a map of S(vrms,t0); (2) manual picking is performed in the semblance map for several stack times t0; and (3) interval velocities, vint, are calculated based on the picked smooth stack velocities, vrms, to construct an earth velocity time model that does not require a reference subsurface model.In conclusion, the present automatic velocity analysis has multiple tasks: (1) diminishing the picking step by considering that the stack velocities are based on an interval velocity model; (2) searching for an interval velocity model that best explains the estimated stack velocities; and (3) automatically searching, subject to geological, physical and mathematical constraints, and editing.