This paper describes the mapping of vegetation patterns in two southern Mongolian protected areas, namely the Gobi Gurvan Saykhan National Park and Great Gobi B Strictly Protected Area. Both areas were mapped individually. Field-sampling followed a modified Braun-Blanquet approach, relevés were classified using phytosociological methods. Training areas were assigned to phytosociological units and were enlarged using various algorithms such as red-green-blue images, NDVIs, Tasselled Cap and unsupervised classifications. The enlarged training regions were used as ground truth data for a supervised classification of Landsat ETM+ scenes. The classifications were performed with a maximum likelihood algorithm and made more homogenous by applying a nearest neighbour filter with a 7×7 pixel range. Accuracy was checked with additional reference data sets taken in the field. The overall accuracy was well above 90%. Misclassifications occurred mainly due to the ecotonal character of some plant communities, or due to the occurrence of mixed pixels. In order to make results accessible for park management and administration, the final data sets were implemented in a geographical information system that includes other thematic information such as country and districts boundaries, wells, rivers, lakes, as well as a high resolution digital elevation model. The mapped areas represent two out of five protected areas in Mongolia that are situated in the southern Gobi. In this paper, they are for the first time compared with respect to their community composition and vegetation distribution.