This paper proposes a novel Multi-Focus Image-Fusion algorithm based on Non-Uniform Rectangular Partition (NURP). In NURP, some bivariate polynomials over partitioned girds are obtained to represent the image pixels. Usually, in the same area, partitioned grid number and size in the clearer sub-image will be greater and smaller than those in blurry sub-image. Therefore, by comparing the partition grid size of the same area in both multi-focus images, the best pixel can be determined as the final fusion pixel. In addition, the control error in NURP will control the partition accuracy, the less the control error, the more partition grids needed to reconstruct the image. Therefore the control error will affect the partition results thus affecting the judgment of the fusion pixel. In order to obtain more correct fusion pixel, the control error will be set to different values to get different partition grids for the multi-focus images. Each fusion pixel in the multi-fusion images will be counted separately and the one with the highest frequency is chosen as the final fused pixel. For the pixel with the same frequency, set the final pixel value to be the mean of the pixel values from multi-focus images. Experimental results prove that this algorithm lets the fused image obtain some better objective evaluation indexes and enhances the spatial detail information of the fused image thus improving its clarity visually.