A stereo matching method for 3-D scenes is described. The problems addressed are a definition of a search space for scanlines of stereo images and a formulation of the determination of the optimal path in the search space. The SSD (Summation of Squared Differences) method is a measure of similarity between a right image region and a left image region and is used as a cost function of the optimal problem. There are some problems regarding the SSD modeling. To resolve the problems we propose a stereo matching method based on the WMDL (weighted minimum description length) criterion in which parameters of the problem are optimized. The WMDL criterion is an information criterion proposed by us previously. The efficiency of the method is shown by using both synthetic and real images. As the results show, 3.1% to about 18.9% of the error values decreased for the synthetic images and 8.07% to about 12.19% of the error value decreased for the real image in comparison with the SSD method.