A snake-based algorithm for segmenting an object from a pair of stereo images is presented. Unlike previously developed snake-based algorithms, this one performs well even when the objects in the picture are occluded and the background behind them is cluttered. Also, the algorithm is not sensitive to the placement of the initial snake points. The algorithm uses a new energy function defined over the disparity space between the pair of the stereo images to successfully locate the boundary of an object in a complex image. Experimental results have shown that the developed algorithm produces more accurate segmentation results than those of the well-known conventional snake algorithm reported by Kass et al.