Abstract: We propose a new method for extracting loess shoulder-lines from grid DEM.The morphological characteristics of loess shoulder-lines are investigated firstly.By applying the edge detection approach,a new method for extracting loess shoulder-line candidate points is proposed based on the prominent height variation of the points.The algorithm then connects the candidate points to small line segments by morphological methods.Finally,precise,systematic loess shoulder-lines are extracted after refining the line segments.Experiments in the loess hill area show that the extracted lines from gradient based operators,such as Sobel,Roberts and Prewitt,have relatively poor results on both integration and matching precision with manually extracted loess shoulder-lines,while loess shoulder-lines extracted by LOG operator has very good matching ratio to the manually extracted loess shoulder-lines.So LOG is an ideal loess shoulder-lines extraction operator and can be used to extract loess shoulder-lines effectively and automatically.