Pangquangou National Nature Reserve is well known as ‘the distribution centre of Prince Rupprecht's larch (Larix principis-rupprechtii)' in China. Community variation in Prince Rupprecht's larch forest provides habitat heterogeneity for animals, especially for the endemic and endangered brown-eared pheasant (Crossoptlon mantchuricum). In this study a total of 120 quadrats (each 10×10m) were established to measure and record species composition and six environmental variables to examine the underlying variables that control community variation. We applied a multivariate regression tree analysis to detect community variation, and used redundancy analysis-based variation partitioning to separate the effects of environmental variables on community variation. The results show that Prince Rupprecht's larch forest in the Pangquangou National Nature Reserve can be represented by eight community types. The amount of total species variability captured by all environmental variables was 20.6%, and the cumulative percentage variance of species–environment relationships was 95.8%. However, analyses with a conditional effect approach revealed that elevation, aspect and litter thickness contribute the most to community variation. The pure and joint effects of these three explanatory variables were separated with variation partitioning analyses. The results highlight that the effect of elevation accounts for the largest fraction of community variation in Prince Rupprecht's larch forest.