EDGE preserving smoothing is an important step toward image segmentation. Currently, bilateral and mean shift filters are popular solutions to this problem. The multiscale mode filter proposed in this paper can be thought of as a generalization of both, working successively at a number of scales. As a result of the multiscale approach, the filter avoids being trapped into spurious local maxima, in favour of more significant ones. Compared with the conventional mean shift filter, the multiscale mean shift filter produces smoother images and restores blurred edges more effectively. The filter can cope well with images corrupted by heavy noise.