Abstract. Studies designed to understand species distributions and community assemblages typically use separate analytical approaches (e.g., logistic regression and ordination) to model the distribution of individual species and to relate community composition to environmental variation. Multilevel models (MLMs) offer a promising strategy for integrating species and community-level analyses. Here, we demonstrate how MLMs can be used to analyze differences in species composition of communities across environmental gradients. We first use simulated data to show that MLMs can outperform three standard methods that researchers use to identify environmental drivers of the species composition of communities, redundancy analysis (RDA), canonical correspondence analysis (CCA), and nonmetric multidimensional scaling (NMDS). In particular, MLMs can separate the effects of collinearity among environmental drivers and factor out the effect of changes in overall species abundances or occurrences that do not involve changes in composition. We then apply MLMs to presence/absence data for 14 species of understory herbs and topographic, biotic, and edaphic variables measured in 54 forested plots in the Southern Appalachian Mountains. In addition to providing information about community composition, MLMs simultaneously identify the responses of individual species to the environmental variables. Thus, MLMs not only have potentially superior statistical properties in analyses of community composition compared to standard methods, but they simultaneously provide detailed information about species-specific responses underlying the changes in community composition.