In this project, we seek to characterize the differences between the tongue movements of normal speakers and those of speakers with cerebellar ataxia, based on the analysis of sequences of ultrasound images. Specifically, the aim is to extract tongue surface features that are both biologically meaningful and statistically useful. To facilitate and automate this process, we developed a special-purpose image processing system for a Macintosh II personal computer. The conception of this system is modular. The following are the successive processing steps: (1) a prefilter for noise reduction; (2) a geometrical transformation (polar coordinates); (3) an enhancement of the tissue-air interface in the surface region of the tongue by matched filtering; and (4) an extraction of border points by searching for an optimal radial path along the angular dimension. The last-mentioned task is performed very efficiently using dynamic programming. This system can handle image sequences in a fully automatic mode. The procedure was tested using a series of representative tongue profiles; the computer-extracted contours were found to be in very good agreement with the manual tracings of an experienced experimenter.