Smoothlets, as one of the adaptive geometrical multi resolution methods, are efficient in the approximatb1ion of images with various edges. However, when multiple edges present in blocks, smoothlets become less efficient due to its requirement in further partitioning the blocks. To overcome this problem, multismoothlets have been developed, which could efficiently represent multiple edges with a vector of smoothlets. However, the improvement in terms of the quality of the reconstructed image by the existing implementations of serial and parallel multismoothlets is relatively small. In this paper, we propose a new residual multismoothlets method by iteratively calculating the smoothlets at the original image and multiple levels of residual images to obtain the multismoothlets. Experimental results demonstrate that the proposed method can effectively improve the quality of the reconstructed image in image approximation compared with the existing multismoothlets.