The extraction of shoulder lines of valleys on the Loess Plateau is the basis of constructing the surficial spatial distribution of soil and water loss. A shoulderline extraction method was proposed by using optimized topographical texture information. Based on the 1∶10000 DEM data and GF-1 remote sensing images, areas with obvious shoulder lines of valleys on the Loess Plateau were taken as the research object. Considering texture features of the gray level cooccurrence matrix of the terrain, along with geometrical structures of ground objects, terrain texture and differences of adjacent pixels, the gully edge of the typical watershed was extracted by using the objectoriented method. The extraction results were further validated by comparing with the morphological characteristics of the slope and the draining characteristics of the catchment. The results showed that elevation, illumination simulation, surface depth of cut, slope and other related weakness, as well as the topography texture features such as homogeneity, variance, contrast, correlation, etc., were employed in the ridgeline extraction. Compared with empirical interpretation based on experts’ knowledge, offsets within four pixel for over 90% of the grid, with the average of absolute errors within 2~3 pixel and the maximum error was 4~6 pixel. The method fully exploited the features of strong correlation and heterogeneity between pixels in the upstream and downstream of the ditch margin, which had strong antinoise ability and effectively weakened the mixing of positive and negative topography units. It also reduced the data redundancy of postprocessing, and realized the balance of classification accuracy, efficiency and operability, which provided an effective method and approach for the macroscopic landform differentiation on the Loess Plateau.