Abstract: With the impletion of the policy of great development of Western China, the process of urbanization in the western China has increased rapidly, especially for some big cities. At the same time, urbanization also affected urban periphery ecological environment deeply. So how to get information about urban residential areas in time and accurately is important to urban monitoring. The study area is in the Weihe Plain, where the precipitation is low and barrens are distributed widely. In this research, to retrieve urban residential areas in the west of China, Radarsat SAR image and Landsat TM image were emerged using HIS transformation and then 2-class supervised classification was used to obtain urban land use information, that were urban areas and non-urban areas. This method takes advantage of both multi-spectral image and radar image. As we know, in the TM image the spectral features of urban residential areas and barrens are so similar that we could not distinguish them. However, the radar images can differentiate urban residential areas from barrens, for SAR data are sensitive to residential areas. To reveal the advantage of this method, the TM image was also classified into 5 land covers, namely urban residential areas, farmland, woodland, water body, and uncultivated land (including barrens) using traditional supervised classification. By comparing the two different methods, we find that the urban residential areas derived from the merged image of TM and SAR is more accurate than that using TM image only, the overall accuracy of them are 84.21% and 71.79%, respectively. So we considered that the image fusion method is an effective way to retrieve urban residential areas for the arid region where barrens are distributed, which could not only obtain all the residential information, but also eliminate barrens, thus the retrieving accuracy is very high.