Elevation of earth surface can be modeled into several forms, i.e. height points, contour, Digital Elevation Model (DEM). DEM as a continuous type of elevation data is generated mainly from height point data. The accuracy of DEM depends on many factors; which one of them is the level of sampling density of the height point used as the input of interpolation processing. Height point itself can be generated from various remote sensing technologies. Stereo optical is very common nowadays in producing DEM data. Spatial resolution or pixel size specification of the stereo optical image might effect on the level of sampling density of height point generated. This research is aimed to analyze the influence of different sampling density generated from some optical stereo images and its pixel size in generating DEM as a model of continuous elevation. The experiment is using three types of optical stereo data which has different sampling density and spatial resolution. The result is shown that the more sampling density the more accurate DEM generated, and the more detail pixel size the more accurate DEM generated. Meanwhile, two data with the same pixel size might significantly produce a different accuracy of DEM. It can be caused by another factor such as input data errors and processing stage errors.