This paper proposes a multi-level vector error diffusion method for reducing smear artifacts in boundary regions. These smear artifacts mainly result from a large accumulation of quantization errors. In particular, smear artifacts appear as color bands the width of a few pixels along transition lines. Therefore, to reduce these artifacts, the proposed method excludes the spreading of a large accumulated quantization error by selective error propagation in the error diffusion process. First, the vector norm of the difference between the error-corrected color vector and the original color vector for neighboring pixels is computed and tested against a threshold. Then, if the first condition is satisfied, the vector norm of the difference between the error-corrected color vector and the nearest one among the 64 primary color vectors is computed and compared to a threshold. Finally, the difference in the vector angle between the error-corrected color vector and the original color vector for neighboring pixels is computed and compared to a threshold. Accordingly, the propagation of a quantization error onto neighboring pixels in the error distribution process is only carried out if the above three conditions are satisfied. As a result, the proposed method effectively decreases the appearance of smear artifacts in boundary regions.