Gurbantunggut desert is the second largest desert in China after Taklimakan desert, with an area of 48,800 square kilometers, of which 97% is covered with fixed and semi-fixed sand dunes. From May to June in 2009, plant diversity was surveyed in 61 sites in the southern Gurbantunggut Desert using a stratified random sampling design. The sampling plot at each site is 1000 m². 92 plants belonging to 71 genera and 22 families were observed. With many families and genera including only one species, the flora dominance is obvious. Among these species, 75 species are herbs, accounted for 81.5% of the total number of species; 40 species are ephemeral plants, accounted for 43.5% of the total number of species, and the cover accounted for 54% of total canopy cover of the desert. The flora analysis showed that most species belonged to Mediterranean, and West Asia to Central Asia (areal-type 12) areal-types, reflecting the arid and semi-arid geographical features of the area. The total number of species, the number of herbs and the number of ephemeral plants correlated negatively with latitude but positively correlated with longitude and altitude. From south to north, total coverage, herb coverage and the coverage of ephemeral plants declined significantly, but the coverage of shrubs increased. From west to east, total coverage, herb coverage and the coverage of ephemeral plants increased, shrub coverage is on the contrary. Multivariate regression trees were used to expound on the relationships between plant communities and 18 environmental variables (including soil, geography and climate). The MRT categorized communities into four groups in terms of latitude, pH of soil and elevation. The community group in higher precipitation had higher species richness. Taking the variation in species richness and coverage caused by precipitation difference in the spatial into consideration, it is predicted that the herbs advantage will be more obvious with increased precipitation in the future.