A method of processing seismic data on parallel processors, preferably on a massively parallel processor. The input ot the process is seismic data from one or more of a number of seismic lines. The end product of the invention is a DMO corrected, zero-offset seismic image of the subsurface. By repeating the method on different input offsets, the variation in reflection strength of a reflector as a function of the angle of incidence can be examined. The method includes a two stage parallelization. First, a parallelization over the incoming traces generates a suite of DMO-corrected partial images for each point on each incoming trace. Secondly, a parallelization over output locations accumulates and combines the partial images, and produces output traces.