Throughout history, humans have tried to represent what they see through images. Mapmakers have always sought ways inwhich to represent both the location and the three dimensional shape of land. At the beginning, the way to obtain a 3D representation of land was to measure planimetry and height (as we can identify later by longitude, latitude and height) using basic measuring devices. Nowadays, the improvements of airborne and spatial instruments make it possible to produce images by sensing the electromagnetic radiation from the Earth. So, we can distinguish two classes of remote sensors: optical sensors and radar sensors. Optical sensors, such as Landsat or SPOT 5, operate around the visible spectrum and provide images with a fine resolution (less than 5 meters for SPOT 5). Thus, these kinds of sensors become very useful for civilian applications (cartography, elevation map, agriculture, hydrography, management of natural hazards, meteorology, geology, deforestation and so on). Considering the subject of this chapter, the extraction of terrain elevation by stereoscopic images can give digital elevation models with an error of about 5 meters (Toutin, 2000). However, optical sensors could be critically useless because of weather conditions or lack of light (i.e. sun). Thus, the use of radar sensors is a good way to overcome the limitations of optical sensors: not very sensitive to rain, considered as active sensors (because they have their own source of energy). Thanks to the signal processing applied to radar signal (pulse compression and synthetic aperture), radar systems can provide images with a very high resolution (for example, Radarsat-2 has an ultra-high resolution mode of about 3 meters for resolution). So, radar images are considered as additional information to optical images. With regard to these properties, one can estimate that radar images are used to get elevation terrain. The more intuitive way to extract depth information from remote sensing images is stereogrammetry. As the brain operates on optical images from eyes, the technique of radargrammetry is applied to SAR (Synthetic Aperture Radar) stereo data and provides digital elevation models (DEM). Considering this preamble to the radargrammetric world, this chapter examines one way to produce digital elevation models (DEM) from a mountainous area (the French Alps) and the way to improve the accuracy of the DEM. So, we will organize the discussion in three parts. In part 1, in order to better understand the stereo computation, we need to explain the basic characteristics of a radar image, which is particularly important to be considered during the radargrammetric processing. Thus, a radar image can be seen as a distribution of reflected electromagnetic energy on the ground. So, each element (i.e. a pixel) of an image is described by its size along the azimuth and range axis. Also, specific characteristics of a radar image are described as layover, shadowing and foreshortening. Because radargrammetric processing is