This work presents a Graphic Processing Unit (GPU) implementation of a pre-stack two-way wave equation migration method known as reverse time migration (RTM). The GPU implementation of the RTM algorithm is based on the finite difference scheme in time and space, to compute the two-way wave propagation. The boundary conditions are based on the Convolutional Perfectly Matched Layer (CPML) strategy. Experimental results of the migrated image for the Marmousi model using the algorithm implementation are shown. In addition, results of the execution time of the RTM algorithm implementation in both architectures, a NVIDIA Fermi GPU and a INTEL CPU, show a speed-up of approximately 4x in the GPU implementation. Furthermore, results of disk transfer vs. RAM memory handling, and granularity show the benefits of the implementation on a GPU architecture.