Abstract: China Loess Plateau is world-wide famous for its peculiar and unique landscape, in which the loess shoulder-lines zigzag on the surface making a distinct separation between loess positive terrain and the negative one. The automatic segmentation of loess positive and negative terrains (P-N terrains)with precise location and high efficiency, which is a crucial technique in constructing the mechanism-process models of loess landform drainage systems, is still need to be further improved. Nowadays, with the increasing availability of digital topographic data, the Digital Elevation Model (DEM)can now be applied to better description of natural topography, and it is also regarded as the most proper data source for automatic extraction of loess positive and negative terrains. In this paper, the conformation and flow confluence characteristic of the hill-slope in representative loess landforms is discussed. Based on this, an auto-segmentation method of P-N terrains from 5 m resolution DEMs is proposed. There are three key procedures in the operation. The first step is the identification of grid points on the shoulder-line position by considering the slope difference up and down. Slope gradient 30°,25°and 20°are used as the thresholds in loess platform area, loess ridge-hill area and loess hill area respectively in the extraction model. The second stage is to expand shoulder-lines’candidate cells directionally by considering the spatial direction of the local hill slope aspect and the trend of the shoulder-lines. This step is helpful for deriving more consecutive and detailed shoulder-lines on DEMs. The third step is the generation of positive terrain which can be regarded as the extraction of the upstream area of the shoulder-line points by using the hydrologic analysis model. All the shoulder-lines’cells derived above are imported into the model as the pour point data. This is an alternative method for evading the difficulty of converting shoulder-line grid cells into consecutive vector lines. Validation tests are took out by contrasting the auto-extraction results from DEMs and the delineating results from 1 m resolution DOMs in six drainages. Results show that maximum area difference between the positive terrain area derived above and actural area is 1.15 km², and the percentage of the cells distance offset values less than 10 m exceed 95%. So the main advantages of this approach are high accuracy, lower demands on manual intervention and ready availability of required data for many regions on the Loess Plateau. The morphology of the earth surface is the interactional production of the hydrosphere, lithosphere, atmosphere and the biosphere, thus the morphology is the external representation, and the interaction process is the inner dynamical force. The extraction of the topographic feature lines not only need considering the characteristic of the terrain morphology, but also need considering the inner mechanism of land surface process.