We develop and implement a computer algorithm to image process (migrate) 3-D prestack seismic reflection data on the massively parallel Connection Machine (CM) architecture. Our Kirchhoff 3-D wave-equation time migration achieves our initial goal of 400 Mflop/s on an 8k processor CM, and is scalable to over 3 Gflop/s on a full 64k processor Connection Machine. This is accomplished by parallelizing the migration problem in surface coordinates (x,y), and by serial looping over the pseudodepth coordinate , thus optimizing the amount of in-processor parallel computation. Data I/O rates between the front-end and the CM are enhanced by first transposing the data and then calling parallel processor data access subroutines. Internal data movement is enhanced by calling indirect memory addressing and circular data shifting subroutines.