A new stereo matching algorithm is introduced that performs iterative refinement on the results of adaptive support-weight stereo matching. During each iteration of disparity refinement, adaptive support-weights are used by the algorithm to penalize disparity differences within local windows. Analytical results show that the addition of iterative refinement to adaptive support-weight stereo matching does not significantly increase complexity. In addition, this new algorithm does not rely on image segmentation or plane fitting, which are used by the majority of the most accurate stereo matching algorithms. As a result, this algorithm has lower complexity, is more suitable for parallel implementation, and does not force locally planar surfaces within the scene. When compared to other algorithms that do not rely on image segmentation or plane fitting, results show that the new stereo matching algorithm is one of the most accurate listed on the Middlebury performance benchmark.