Recently, video lossless compression has been developed for applying it to digital cinema, video archiving of contents, etc. Video lossless compression is important in image processing problem since a large image requires a large amount of storage space. The purpose of this paper is to enhance the predictor used for the lossless compression of video. In this paper, we propose the 3-dimensional predictor for the effective prediction. In addition, the three-dimensional spatio-temporal gradient is adopted to improve the conventional image compression methods such as GAP, MED which are two-dimensional predictions based on horizontal and vertical gradients. The spatio-temporal gradient is a spatial data resulted from the projection of triangular prism composed of the neighborhood pixels. From the experimental results compared with the previous prediction methods, we confirmed that the prediction using proposed method is more efficient.