The object-oriented classification approach was employed in loess hilly-gully region. Taking typical watershed Yan’gou as example, ALOS satellite images were used. The object-oriented multi-segmentation was carried out with multi-spectral, panchromatic imagery, auxiliary data of digital evaluation model (DEM) and the normalized difference vegetation index (NDVI). The land-use categories closely related to ecosystem restoration, farming and living, such as forest, grass, farmland, orchard, settlement and water, were classified with thresholds. The classification results had promising accuracies, and the overall classification accuracy was 77.73%.