Landforms can be described and identified by parameterization of digital elevation model (DEM). This paper discusses the large-scale geomorphological characteristics of China based on numerical analysis of terrain parameters and develop a methodology for characterizing landforms from DEMs. The methodology is implemented as a two-step process. First, terrain variables are derived from a 1-km DEM in a given statistical unit including local relief, the earth's surface incision, elevation variance coefficient, roughness, mean slope and mean elevation. Second, every parameter regarded as a single-band image is combined into a multi-band image. Then ISODATA unsupervised classification and the Bayesian technique of Maximum Likelihood supervised classification are applied for landform classification. The resulting landforms are evaluated by the means of Stratified Sampling with respect to an existing map and the overall classification accuracy reaches to rather high value. It's shown that the derived parameters carry sufficient physiographic information and can be used for landform classification. Since the classification method integrates manifold terrain indexes, conquers the limitation of the subjective cognition, as well as a low cost, apparently it could represent an applied foreground in the classification of macroscopic relief forms. Furthermore, it exhibits significance in consummating the theory and the methodology of DEMs on digital terrain analysis.