Stereo matching is one of the most popular techniques to estimate dense depth\nmaps by finding the disparity between matching pixels on two, synchronized and\nrectified images. Alongside with the development of more accurate algorithms,\nthe research community focused on finding good strategies to estimate the\nreliability, i.e. the confidence, of estimated disparity maps. This information\nproves to be a powerful cue to naively find wrong matches as well as to improve\nthe overall effectiveness of a variety of stereo algorithms according to\ndifferent strategies. In this paper, we review more than ten years of\ndevelopments in the field of confidence estimation for stereo matching. We\nextensively discuss and evaluate existing confidence measures and their\nvariants, from hand-crafted ones to the most recent, state-of-the-art learning\nbased methods. We study the different behaviors of each measure when applied to\na pool of different stereo algorithms and, for the first time in literature,\nwhen paired with a state-of-the-art deep stereo network. Our experiments,\ncarried out on five different standard datasets, provide a comprehensive\noverview of the field, highlighting in particular both strengths and\nlimitations of learning-based strategies.\n