Introduction of the metacommunity concept has made a substantial contribution to better understanding of community composition and dynamics of a fragmentized habitat at a regional level. Leibold and Miller defined metacommunity as a set of local communities that are linked by dispersal of multiple potentially interacting species. The metacommunity concept has provided a new and revolutionary paradigm for the community ecology by emphasizing integrated variation among communities at the regional level. This variation is accentuated by biotic interactions and spatial variation in environmental characteristics and modulated by dispersal among sites. The basic problems of metacommunity research are the multiple species coexistence mechanism, the reasons behind the diversity, and mechanism maintaining this diversity within the same system. Leibold and Miller proposed four paradigms for metacommunities: the patch-dynamic view, the species-sorting view, the mass effects view, and the neutral view, each emphasizing different processes of potential importance in metacommunities. A variety of studies verified the four ecological paradigms, but only few addressed the terrestrial vertebrates system. In recent years, as a result of human activities and natural disturbances, Alashan Desert habitat in Inner Mongolia has been largely fragmentized. This habitat fragmentation is a major threat to animal communities. From 2010 to 2012, we investigate rodent communities and habitat environmental factors at eight fixed experimental sites in Inner Mongolia's Alashan Desert. We used a set of full and partial redundancy analyses to estimate four additive partitions of variance in rodent species composition: (a) unexplained variation, (b) variation explained by spatial characteristics, (c) variation explained by local environmental conditions, and (d) variation explained by both spatial and environmental characteristics. The results of full and partial redundancy analyses showed that habitat and spatial characteristics alone accounted for 72.8% and 33.8%, respectively, of the variation among sites regarding rodent species composition. All explained variables accounted for approximately 86.5% of the variation (P = 0.032) in rodent species composition among sites. Spatial characteristics accounted for 13.7% of the variation (P = 0.246) in species composition after removing variation shared with habitat characteristics. Habitat characteristics accounted for 52.7% (P = 0.016) of the variation in species composition after removing variation shared with spatial characteristics effect. Finally, spatial structured habitat characteristics (interaction term) accounted for approximately 20.1% of the total variation in rodent species composition among sites. The environmental component to rodent species composition was greater than the spatial component. Pure environmental characteristics accounted for significant amounts of variation in rodent species composition, and pure spatial characteristics accounted for non-significant amounts of variation in rodent species composition. The results from variance decomposition suggested that the species sorting model of metacommunity structure fits better the rodent community than the mass effect or neutral models.