In conventional seismic processing, velocity analysis is performed by using the normal moveout (NMO) equation which is based on the assumption of flat, horizontal reflectors. Imaging by migration (either before or after stack) is done normally in a subsequent step using these velocities. In this paper, velocity analysis and imaging are combined in one step, and migration itself is used as a velocity indicator. Because, unlike NMO, migration can be formulated for any velocity function, migration-based velocity analysis methods are capable of handling arbitrary structures, i.e., those with lateral velocity variations.In the proposed scheme, each shot gather (profile) is migrated with an initial depth-velocity model. Profile migration is implemented in the (x, omega ) domain, but the actual implementation of profile migration is not critical, as long as it is not done in a spatial-wavenumber domain, which would preclude handling of lateral velocity variations. After migration with an initial velocity model, the velocity error is estimated, and the initial velocity model is updated; the process is repeated until convergence is achieved. The velocity analysis is based on the principle that after prestack migration with the correct velocity model, an image in a common-receiver gather (CRG) is aligned horizontally regardless of structure. The deviation from horizontal alignment is therefore a measure of the error in velocity. If the migration velocity is lower than the velocity of the medium, events curve upward, whereas if the migration velocity is higher than the velocity of the medium, events curve downward.