A digital elevation model (DEM) presents immense data proving three dimensional terrain structure of any part of the Earth. DEMs are obtained by two main methods in space-borne remote sensing as stereoscopy using optical or radar imagery and interferometric synthetic aperture radar (InSAR) technology. In fact, the primary product of space-borne remote sensing techniques is a digital surface model (DSM) that contains points located on the top of ground objects. By removing these points that do not belong to the bare ground, the DEM is obtained. In optical imagery, DSMs are generated based on stereo matching using ground control points and co-located clear tie points at stereo image-pair with high correlation. In this case, correlation comes into prominence and affects the success of DSM acquired by stereoscopy. This investigation aims to assess the quality of DEMs produced by medium resolution spatial data derived from optical imagery depending upon the effect of correlation in stereoscopy correspondingly the land cover types. Towards this purpose, land cover classes have been generated such as open, forest, built-up, road network and rocky regions, DSM-DEM conversion was applied by optimal filtering methods and DEM accuracies have been achieved separately. The analyses were realized using actual ASTER GDEM Version 2 with 30m original grid spacing in Zonguldak, Turkey including rugged topography and suitable land cover classes. For the verification, a reference DEM derived from 1/1000 scaled aerial photos was employed.