To improve the accuracy of machine vision based maize leaf disease segmentation under the background of farmland, a graph theory based approach to maize leaf disease segmentation in multi-color space was proposed. The graph theory was used in different color spaces and the maize leaf disease segmentation was then formulated as a graph segmentation problem in different color spaces. Furthermore, an effective fusion method was applied to update the initial segment result. Experiments on the maize leaf disease showed that the segmentation results were good. The results revealed that the proposed approach was feasible and effective.