Recent developments in open source digital elevation models (DEMs) increase their applications in different disciplines. The lack of proper information about the accuracy of these models in regions with different topographic characteristics, make their analysis unreliable. Most estimates of accuracy, e.g., mean, standard deviation and RMSE depend on an implicit assumption that the errors comprise a random sample from a normal distribution. But, analytic data often depart from that assumption. Therefore, this paper first study statistical characteristics of SRTM and GDEM open source global DEMs. Then we define a confidence interval for RMSE which make it possible to investigate the appropriateness of reference points. Afterwards, measures for accuracy assessment of DEMs based on robust methods in L1 norm and Huber's method as well as conventional methods in L2 norm are discussed. In order to consider different topographic characteristics we compare SRTM and GDEM based on the defined measures in Urban, Flat, Hill and Mountain regions. Finally, we studied how the errors propagate into slope and aspect maps. It was found that although GDEM has better resolution, but SRTM performs better based on the defined measures in all the regions. It is also illustrated that the accuracy of both models in urban are better than other regions.