Traditional stereo systems often falter over changes in lighting between their two views. Unfortunately, such changes often occur when using stereo with a wide baseline or between images from different spectra. In this paper, we propose a new dense stereo correspondence similarity metric, mutual information, which has the potential to overcome such adverse conditions. We explore the strengths and weaknesses of this metric, both quantitatively and qualitatively, under a variety of conditions. Throughout the exploration, we compare mutual information to a more traditional cross-correlation stereo system. We show that mutual information performs under conditions in which traditional dense stereo fails.