1. Overyielding in mixed-species forests has been demonstrated in a vast body of literature, and the focus of functional biodiversity research is now shifting towards a mechanistic understanding of these observations. 2. We explored diversity-productivity relationships (DPRs) at two sites of a large-scale tree diversity experiment, with benign (Zed) and harsh (Ged) environmental conditions for plantation establishment. Additive partitioning methodologies were adopted to detect phenomenological patterns in the productivity data, and the trait structure of mixed communities was used to advance insights into compositional effects. 3. After six years of plantation development, biomass productivity was significantly higher in mixtures compared to the monocultures of component species. We observed that processes operated through direct tree-tree interactions, since the diversity signal disappeared where trees in mixed stands were surrounded by conspecific neighbors only. This result is particularly relevant for mixed-species planation systems, as trees are commonly planted in monospecific patches to simplify the management. Partitioning unveiled strong selection effects at both plantation sites. However, at the Ged-site this was caused by competitive dominance of species with fast young growth whereas at the Zed-site, species with slow young growth improved their performances but not at the expense of others (i.e. trait-dependent complementarity). Species tolerance to shading is an influential trait to predict biodiversity effects, with community-weighted means in shade tolerance mediating dominance effects (Ged) and functional diversity in shade tolerance mediating (trait-dependent) complementarity effects (Zed). 4. Synthesis. This study highlights that biodiversity effects in young tree plantations could be explained by the functional composition of mixed communities, with a key role for species levels of shade tolerance. As contrasting results between plantation sites were observed, future research should target the context-dependency of DPRs.