We measured a diverse range of foliar characteristics in shrub and tree species in temperate rainforest communities along a soil chronosequence (six sites from 8 to 120 000 years) and used multilevel model analysis to attribute the proportion of variance for each trait into genetic (G, here meaning species-level), environmental (E) and residual error components. We hypothesised that differences in leaf traits would be driven primarily by changes in soil nutrient availability during ecosystem progression and retrogression. Several leaf structural, chemical and gas-exchange traits were more strongly driven by G than E effects. For leaf mass per unit area (MA), foliar [N], net CO2 assimilation and dark respiration rates and foliar carbohydrate concentration, the G component accounted for 60–87% of the total variance, with the variability associated with plot, the E effect, much less important. Other traits, such as foliar [P] and N : P, displayed strong E and residual effects. Analyses revealed significant reductions in the slopes of G-only bivariate relationships when compared with raw relationships, indicating that a large proportion of trait–trait relationships is species based, and not a response to environment per se. This should be accounted for when assessing the mechanistic basis for using such relationships in order to make predictions of responses of plants to short-term environmental change.