Urban land use/cover mapping is very important and it is the base of further urban analysis and research. Whereas urban land use/cover mapping of using remotely sensed images having medium spatial resolution presents numerous challenges due to the intensive heterogeneity of urban landscapes. In order to solve the above challenges and improve the accuracy of urban land cover/use mapping, we proposed a novel approach. Firstly, fraction image is attained based on spectral mixture analysis, and normalized MNF image is gained based on spectral normalization of the origin image and MNF transform. Secondly, combination image is produced based on fraction image and normalized MNF image. Finally, we performed decision tree classification to the combination image and gained urban land use/cover mapping. An ETM+ image acquired in 2001 was used as data source and Nanjing City, China was selected as study area. The accuracy of classification result was validated using IKONOS images of the study area acquired in 2000 and was compared with the other three classification schemes. Results show that this decision tree classification scheme based on the combination image of fraction image and normalized MNF image is superior to the other classification schemes evidently.