Landslides are generally not well observed on single SAR satellite images. On the example of the Chicamocha valley in Colombia, which served as a test zone for this work we show that stereoscopic observation of stereo-simulated ERS-1 or JERS-1 images is efficient. The simulation consists of generating orthoimages using a digital elevation model (DEM) producing vertical and oblique views. Landslides are detected due to the stereoscopic perception of small characteristic geomorphic features. The JERS-1 data are favoured over ERS-1 data because of their greater incidence angle.