Aim To quantify how mammal community structure relates to heterogeneity of vegetation for palaeoecological reconstructions, and to test whether historical or environmental factors are more important in structuring communities. Location Sixty-three natural protected areas in Asia, Africa and South and Central America. Methods We defined faunal communities by allocating species to ecological guilds and calculating proportional representation within each guild. Vegetation heterogeneity for each natural protected area was calculated from satellite images. The relationship between these ecospaces was calculated using canonical correlations analysis, redundancy analysis and principal components analysis. We expected that large, herbivorous mammals would be most strongly correlated with open areas. Convergence was tested by independently eliminating the effects of geography and vegetation heterogeneity on the structure of the mammal communities. We expected that vegetation would more strongly structure communities than geographical position. Results We show that the guild structure of communities across habitats is significantly correlated with vegetation heterogeneity. The highest correlation was between small, scansorial-arboreal secondary consumers and heavy tree cover. The first convergence analysis shows American communities distinguished from Asian and African communities; these latter communities show a remarkable convergence in structure. Historical factors only affected the continent whose mammals had experienced a long period of isolation. The second convergence analysis shows that almost all biomes have the same or very similar community structure regardless of continent. Main conclusions Communities from the same environments in different continents showed remarkable convergence. Communities from the same continents only converged when those continents shared a recent geological and biological history. These results suggest that historical and environmental factors are operating over different timescales. This study confirms that environmental reconstructions made on the basis of whole communities will accurately reflect the environment that the community lived in. However, reconstructions made for fossil sites in deep time need to take historical factors into consideration. Small, arboreal and scansorial secondary consumers show the strongest correlation with vegetation, correlating with continuous tree canopy cover. This relationship allows simple reconstructions of the amount of tree cover occurring in a landscape from the proportion of species from the community falling in this ecological guild.