Forest restoration increases global forest area and ecosystem services such as primary productivity and carbon storage. How tree species functional composition impacts the provisioning of these services as forests develop is sparsely studied. We used 10-year data from 478 plots with 191,200 trees in a forest biodiversity experiment in subtropical China to assess the relationship between community productivity and community-weighted mean (CWM) or functional diversity (FD) values of 38 functional traits. We found that effects of FD values on productivity became larger than effects of CWM values after 7 years of forest development and that the FD values also became more reliable predictors of productivity than the CWM values. In contrast to CWM, FD values consistently increased productivity across ten different species-pool subsets. Our results imply that to promote productivity in the long term it is imperative for forest restoration projects to plant multispecies communities with large functional diversity.