Abstract: Improving motion estimation is essential for H.264 system.To speedup the full search for motion estimation H.264 encoding process,a framework is proposed in the paper for parallel processing on GPU using computing unified device architecture(CUDA).The proposed method can the advantage of GPU's powerful computing resources and optimal memory hierarchy as well as the potential parallelism in CUDA.Experimental results show that our implementation on nVidia's GTX280 demonstrates substantial improvement upto 70 times speedup than CPU counterpart available.