Abstract: A region growing model under the framework of MRF is proposed for urban detection and the basic unit of the model is the over segmentation region. Firstly, the texture analysis is employed to obtain the initial seed points. Then Mean Shift algorithm is employed to get the over segmentation and the regions that include seed points are used as seed regions. At last, starting from the seed regions, the finally result of urban is detected through a growing criterion under the framework of MRF. The experiments of QuickBird and Ikonos demonstrate that the model can effectively detect the urban area from the remote sensing images. The average ratio of urban detection is 84.35% and the average ratio of correct detection is 93.16%.