One of the main goals of optical 3D mapping satellites, such as SPOT-5, Cartosat-1, ALOS, ZY-3 and similar satellites is the generation of large ortho and digital surface model mosaics. The direct georeferencing performance of the current satellites is worse than the resolution of their data. This leads to geometric errors between adjacent scenes, if they are processed independently and without ground control. These errors can be reduced or eliminated by georeferencing the images with high quality GCPs or with a block triangulation. A complete coverage of larger areas with high resolution satellite imagery usually requires several months or years of acquisition. The images typically contain strong differences due to different seasons, cloud cover and other effects. This paper introduces our automatic image orientation procedure for satellite images, which can deal with these effects and automatically orient large blocks of satellite imagery. We additionally exploit DSMs to minimize the number of ground control points required for orientation of the images. The method evaluated on a block of 1210 Cartosat-1 images covering northern Italy.