To solve nonlinear seismic velocity inversion problems, we have developed an automatic velocity-estimation technique based on the stochastic method called multiobjective evolutionary algorithms (MOEA). Semblance and differential semblance are used as objective functions. To cope with the high computational cost, we customized MOEA, added domain knowledge (velocity increases with depth, slowly varies along layers, and so forth), which improves the conditioning of the problem and accelerates convergence. This approach is robust because it can cope with large velocity errors. Computational cost of this algorithm is at least two orders of magnitude faster than other stochastic methods and comparable to that of direct gradient methods.