Foundation species are species that play important roles in structuring ecological communities. Therefore, conservation managers often aim to promote foundation species. However, it can be unclear which features of foundation species ought to be the focus of management. We studied hummock-forming grasses in the genus Triodia. Triodia grasses are considered foundation species because they create complex structures used by many fauna species. Although conservation managers often aim to promote extent cover of Triodia, this is only likely to be optimal for species most strongly associated with extent cover or other structural features strongly correlated with extent cover. We tested (i) whether ‘extent cover’ is the most appropriate way to characterise Triodia as habitat and (ii) whether fauna are associated with any non-Triodia structures. We studied the Triodia structure associations of one mammal, two birds and five reptiles associated with Triodia scariosa at 524 sites in the Murray-Mallee, Australia (Ningaui yvonneae, Amytornis striatus, Stipiturus mallee, Ctenophorus fordi, Ctenotus atlas, Ctenotus inornatus, Delma australis and Delma butleri). We used site-level presence–absence data and vegetation structure data to compare parsimony of models built using four Triodia structural features: extent cover, mean height, mean width and mean volume. We also included non-Triodia vegetation structures in the model selection: extent cover of leaf litter, shrubs and trees. We divided structural features related to extent cover into categories according to their heights. One species was most closely associated with mean Triodia height; one species with mean Triodia width and six species with extent cover of Triodia, although here, Triodia height categories differed between species. Five species were also associated with shrubs or trees. Extent cover of Triodia was generally an appropriate measure of Triodia structure. Nevertheless, we found variation between species. When characterising the structure of foundation species, we recommend testing faunal associations with multiple structural features.