Environmental filtering prevents species without certain attributes from occurring in local communities. Traits respond differently to different abiotic factors, assembling communities with varying composition along environmental gradients. Here, we measured proxies of soil fertility, disturbance by fire, response and physiological traits to assess how these variables interact to determine woody species richness and density in a Neotropical savannah. We explicitly incorporated our assumptions about how different abiotic filters influence different subsets of traits into a statistical model using structural equation modelling, yielding a more accurate representation of the assembly process. Fire had an effect on resistance traits, whereas soil fertility influenced physiological traits. Resistance traits explained both the richness and density of plots, whereas physiological traits explained only the density. Fewer fire events led to richer and denser plots. Similarly, areas with lower cation exchange capacity assembled less dense communities. Furthermore, we showed that structural equation modelling yielded a realistic representation of the bivariate interactions of distinct environmental filters with different subsets of traits.