Abstract: Toboggan algorithm is an important tool to image segmentation, and the result of image segmentation depends on how to compute the gradient image to a great extent if applying toboggan to it. A multiscale morphological gradient toboggan algorithm is proposed to obtain the gradient image by characterizing the structural elements of different sizes and then segment the image. To reduce the over-segmentation of toboggan, a merging method of regions is carried out after toboggan segmentation according to their areas and similarities. Test results show that the algorithm is valid and more accurate than other conventional algorithm in image segmentation, especially it has a wider scope of application because of the selective flexibility of structural element shape and size.