Abstract To determine which types of plant traits might better explain ecosystem functioning and plant evolutionary histories, we compiled 42 traits for each of 15 perennial species in a biodiversity experiment. We used every possible combination of three traits to cluster species. Across these 11,480 combinations, clusters generated using tissue %Ca, %N and %K best mapped onto phylogeny. Moreover, for the 15 best combinations of three traits, 82% of traits were chemical, 16% morphological and 2% metabolic. The diversity‐dependence of ecosystem productivity was better explained by the %Ca, %N and %K clusters: compared to adding a new species at random, adding a species from an absent cluster/clade better‐explained gains in productivity. Species number impacted productivity only when all clusters were present. Our results suggest that tissue elemental chemistry might be more phylogenetically conserved and more strongly related to ecosystem functioning than commonly measured morphological and physiological traits, a possibility that merits exploration.