The conventional method for migration velocity analysis scans over one (or possibly multiple) move-out parameter(s) ang the offset or angle axis. The appropriate move-out parameter is then selected based on what produces the best coherence. The coherence as a function of the scanning parameter provides important information on how confident we are of our given move-out measure. In this paper, the move-out parameter selection is set up as an inverse problem and multiple, equi-probable move-out fields are generated. These various realization of move-out are then used to update the velocity model and migrated image. Early results are promising.