Abstract: Two perspective images of single scene taken by uncalibrated perspective cameras are constrained by the epipolar geometry, which is the key to many problems of computer vision. The problem of robust fundamental matrix estimation employing a new method is addressed based on genetic algorithm using different strategies. The method uses each gene to stand for a pair of correspondences, takes every chromosome as a minimum subset for epipolar geometry estimation, and computes the fundamental matrix according to the length of the chromosomes. The method would eventually converge to a globally optimal solution and is relatively unaffected by the outliers. Experiments with both synthetic data and real images show that the method is more robust and precise than other typical methods because it can efficiently detect and delete the bad corresponding points, which include both bad locations and false matches.