An effective aproach based on normalized cuts to segment infrared object was presented. The method is related to the graph theory. An image was represented as a weighted undirected graph, where the nodes of the graph were the pixels of the image, and an edge was formed between every pair of pixels. The weight on each edge was function of the similarity between the two pixels, and the similarity was related to the gray level and the position of pixels. Normalized cut was chosen as the principle to distingusish object from background. The time and space complexity of the algorithm was studied. Four kinds of domain-weight-matrix were defined and a quick table-search method was presented to improve the real-time algorithm. The experimental results show that the proposed method has excellence performance, and it is superior to the traditional segmentation method.