Urban regions are extremely heterogeneous and complex, and include different types of landscapes, which are mosaics of biological and physical patches. Traditional landscape classification approaches based on remote sensing imagery are inadequate to depict the hierarchical characteristics of urban regions. For example, it cannot differentiate between developed land within forest landscapes and that in urban landscapes. To capture the differences, a multi-level classification system is required. Here, we proposed a two-level classification approach based on landscape types and landscape elements, using the Beijing metropolitan region as a case study. In the study area, we identified three landscape types, namely urban, agricultural, and forest, all of which included four types of landscape elements: vegetation, bare soil, water, and impervious surface; forest landscape was without water. The two-level classification was implemented with an object-based method and followed a top-down approach. We first conducted classification of landscape types, and then classified landscape elements separately for each landscape type. With the object-based classification approach, we first segmented the image into objects, and then classified the objects into different classes using supervised classification based on support vector machines (SVM). This process was similar to the way in which human brain perceives landscape and can potentially generate classification with higher accuracy than the pixel-based approach. The overall accuracy of the classification of landscape types and landscape elements was 93.36% and 87.89%, respectively. The misclassification of landscape types occurred mostly in places where urban landscapes mixed with agricultural or forest landscapes. The study area was dominated by urban landscapes, with proportional coverage of 43.54%. The proportions of agricultural and forest landscapes were 36.02% and 20.44%, respectively. As for landscape elements, the urban region was dominated by impervious surfaces, with the proportion of 45.08%. The proportion of impervious surface, however, varied greatly by different landscape types. Thus, it was 70.95% in urban landscapes, but only 38.87% in agricultural landscapes and 12.34% in forest landscapes. Vegetation covered 37.42% of the study area, but only 22.75% in urban landscapes, much less than that of 80.28% in forest landscapes. The percentage cover of water was relatively low (1.99%) in the study area, mostly occurring in the agricultural (56.68%) and urban landscapes (43.32%). Bare soil occupied 15.51% of the whole area, with 76.84% in agricultural landscapes, only 14.28% in urban landscapes, and 8.88% in forest landscapes. Compared to the one-level landscape classification, the multi-level classification explored more information about the relationship between landscape types and landscape elements. Therefore, the multi-level classification method can better characterize landscapes in urban regions and provides a new perspective to linking ecological structure to function.