Simultaneous Algebraic Reconstruction Technique (SART) is able to generate Computed Tomography (CT) images with higher quality compared to Filtered Back-Projection (FBP) method when the projection data is incomplete or noisy. However, it is very time-consuming; and parallel computation is one of those efficient approaches to manage the problem. In this study, a new parallel implementation of SART based on the platform of Compute Unified Device Architecture (CUDA) was proposed. The experimental results show that there are no differences between the images reconstructed by this new method and those by serial implementation, but the reconstruction time is greatly decreased, more applicable to clinical application.