This paper describes a new variational method for estimating disparity from stereo images. The stereo matching problem is formulated as a convex programming problem in which an objective function is minimized under various constraints modelling prior knowledge and observed information. The algorithm proposed to solve this problem has a block-iterative structure which allows a wide range of constraints to be easily incorporated, possibly taking advantage of parallel computing architectures. In this work, we use a Total Variation bound as a regularization constraint, which is shown to be well-suited to disparity maps. Experimental results for standard data sets are presented to illustrate the capabilities of the proposed disparity estimation technique.