There is strong spatial correlation among the adjacent pixels in the image feature area whereas no significant sample correlation exists in the noisy area. Therefore, this paper proposes a novel noise reduction algorithm based on direction correlation of pixels in 8 times 8 block. In the method, four direction gradient values are computed respectively for the block and the orientation correlation characteristic is confirmed according to the maximal and minimum gradient values. Then, we select different templates to filter the block for the different orientation feature. The experimental results show that this method can remove the image noise effectively with low computational complexity, and preserve the original image information simultaneously.