Multitemporal digital terrain models (DTM) are an important source for many purposes such as the detection of areas, which are susceptible to natural hazards such as landslides and glacial lake outburst floods, or for the examination of changes in glacier thickness. To exploit the potential of stereo satellite and aerial imagery for time series analysis, the employed methodology and software can be critical. A statistical analysis based on quartiles is presented to eliminate the influence of registration and elevation errors in DTMs. For our analysis, we used multi-temporal airborne and spaceborne stereoscopic images. The oldest images were recorded in the 1960s by the US American reconnaissance satellite Corona, while the most recent imagery are 2007 Cartosat-1 stereo scenes, along with one ASTER stereo pair. Complex panoramic distortion and limited spatial resolution resulted in the Corona and ASTER DTMs having the highest RMSEz. Due to differing acquisition techniques, applied software packages and temporal differences DTMs will never be identical. Therefore, we propose a relative vertical accuracy assessment with a master DTM. We chose the Cartosat-1 DTM as it showed the highest absolute accuracy. Inaccuracies between the master and the slave DTMs were adjusted by means of trend surfaces and outliers were successfully eliminated applying the interquartile range.