Forests play a vital role in regulation of the global carbon cycle. Mechanistically understanding how their ecosystem functioning relates to biodiversity is necessary for predicting the consequences of biodiversity loss and for setting conservation priorities. Here, we test whether carbon stocks in a subtropical evergreen broad-leaved forest in China are more strongly influenced by plant functional diversity (FD), as would be predicted by the ‘niche complementarity hypothesis’, or by community-weighted mean (CWM) functional trait values, as would be predicted by the ‘mass ratio hypothesis’. Using data from a 24-ha plot subdivided into 400 m2 quadrats, we determined relationships of aboveground carbon (AGC) and topsoil (1–10 cm) organic carbon (SOC) to topographic variables, stem density, CWM and FD of six functional traits hypothesized to influence carbon stocks. After accounting for topographic variables and tree stem density, boosted regression tree models revealed that CWMs were the dominant driving factors for both AGC and SOC, whereas FD had negligible effects. AGC and SOC were influenced by different functional traits, with AGC responding most strongly to CWM values for wood density and maximum tree height, and SOC responding most strongly to elevation, indicating that these carbon stocks are shaped by different underlying mechanisms. Our results support the mass ratio hypothesis but not the niche complementarity hypothesis. Our study implies that, when it comes to maximizing forest carbon storage, conservation priorities should focus on protection of species with traits associated to high carbon stocks.