Tagged magnetic resonance (MR) imaging makes it possible to image the motion of tissues such as the muscles found in the heart and tongue. The harmonic phase (HARP) method largely automates the process of tracking points within tagged MR images. It works by finding spatial points in successive images that retain the same two harmonic phase values throughout the entire image sequence. Given a set of tracked points, many interesting and useful motion properties such as regional displacement or rotation, elongation, strain, and twist, can be computed. When there is a large motion between successive image frames, HARP tracking can fail, and this results in mistracked points and erroneous motion estimates. In this paper, we present a novel HARP refinement method based on seeded region growing that addresses this problem. Starting from a given seed point which is determined by the user to be correctly tracked throughout the entire sequence, this method can reliably track the motion of the whole tissue. A novel cost function is used in the region growing to assure that points that can be most reliably tracked are tracked first. Experimental results on tagged MR images of the tongue demonstrate very reliable bracking.