A new computer algorithm is described by which velocity estimates can be derived from three‐dimensional (3-D) multifold seismic data. The velocity estimate, referred to as “imaging velocity,” is that which best describes the diffraction hyperboloid due to a scatterer. The scattering center is best imaged when this velocity is used in the reconstruction process. The method is based on the 3-D Kirchhoff summation migration before stack. The implementation consists of two basic phases: (1) differentiating the input field traces and resampling them to a logarithmic time scale, and (2) shifting, weighting, and summing each resampled trace to a range of depth levels also chosen on a logarithmic scale. Peak amplitudes in the resulting image matrix give a time T and depth Z from which velocity is obtained using the relation [Formula: see text] The locus of constant velocity is a slanted straight line in the coordinate system of the matrix. In the usual application of migration for velocity analysis, each input trace of N samples is migrated for each of M constant velocity functions requiring [Formula: see text] moveout shift calculations. In the new method presented here, a constant shift is calculated for a given resampled trace, for each depth into which it is summed. This reduces the number of calculations per trace to about N, resulting in a significant improvement in computing efficiency. The operation of the algorithm is illustrated using synthetic and physical model data.