Understanding what drives biodiversity patterns across scales is the central goal of ecology. Both environmental gradients and spatial landscape structure have been found to be important factors influencing species distributions and community composition, and partly reflect the balance of underlying deterministic and stochastic community processes. In some systems, environmental gradients and spatial connectivity are intertwined in that steep environmental gradients serve as boundaries on species movements and impose environment-derived complex spatial structure to metacommunities. Mountainous landscapes are prime examples of this, and recent theory has linked principles of geomorphology, environmental gradients, and spatial structure to make predictions for resulting community patterns. In this context, we examine variation in taxonomic and phylogenetic ant diversity patterns along a geographic transect spanning >5000 m in elevational range in the Hengduan mountains of southern China. We found that environmental gradients dominate variation in both alpha and beta diversity in this landscape, with alpha diversity strongly declining with elevation and beta diversity driven by elevational differences. However, within an elevational band spatial connectivity predicts beta diversity better than geographic distance. Our findings deviate from theoretical predictions in several ways, notably alpha diversity is monotonically declining and within-band beta diversity is invariant with increasing elevation. The discrepancies between theory and observation may be explained by differences in the Hengduan landscape from idealized fluvial landscapes, such as a lack of a mid-elevation peak in connectivity, as well as evolutionary limits on the source pool of species available to populate metacommunities at different elevations. The latter is supported by variation in phylogenetic community structure with elevation. Our results demonstrate the power of conceptual, statistical, and theoretical frameworks that integrate the roles of environment and spatial structure in metacommunities, but that additional work is needed to bridge the gap between abstract theory and real systems.