Stereo correspondence can be formulated as an optimization problem. In this formulation, however, most of the existing solutions adopt gradient-based approaches, whose performance is dependent on the initialization. This paper presents a genetic-algorithm-based solution that is not gradient-based and thus should have less sensitivity toward the quality of the initialization. A specific coding design is employed that represents each solution candidate for the three-dimensional description of the imaged scene as an individual that embraces numerous chromosomes. Through a set of specially designed genetic operators, a population of such individuals is allowed to evolve to reach a globally optimal or near-optimal solution. Our solution scheme also includes a coarse-to-fine search strategy to reduce the matching ambiguity and the computations needed. Experimental results on synthetic and real images illustrate the performance of the approach.