The influence of species diversity on ecosystem functioning has been the subject of many experiments and remains a key question for ecology and conservation biology. However, the fact that diversity cannot be manipulated without affecting species composition makes this quest methodologically challenging. We used partial Mantel tests to evaluate the relative importance of diversity and of composition on biomass production. Here, we applied partial Mantel tests (controlling for the other variable) on two datasets, the Jena (2002–2008) and the Grandcour (2008–2009) Experiment. In both experiments, plots were sown with different numbers of species to unravel mechanisms underlying the biodiversity-ecosystem functioning relationships. Contrary to Jena, plots were not mowed nor weeded in Grandcour, allowing external species to establish. Based on the diversity–ecosystem functioning and competition theories, we tested two predictions: 1) the contribution of composition should increase with time; 2) the contribution of composition should be more important in non-weeded than in controlled systems. We found support for the second hypothesis, but not for the first. On the contrary, the contribution of species richness became markedly more important few years after the start of the Jena Experiment. This result can be interpreted as species complementarity, rather than intraspecific competition, being the driving force in this system. Finally, we explored to what extent the estimated relative importance of both factors could vary when measured on different spatial scales of the experiment (in our case, increasing the number of plots included in the analyses). We found a strong effect of scale, suggesting that comparisons between studies, and more generally the extrapolation of results from experiments to natural situations, should be made with caution.