An enormous recent research effort focused on how plant biodiversity (notably species richness) influences ecosystem functioning, usually through experiments in which diversity is varied through random draws of species from a species pool. Such experiments are increasingly used to predict how species losses influence ecosystem functioning in ‘real’ ecosystems. However, this assumes that comparisons of experimental communities with low vs high species richness are analogous to comparisons of natural communities from which species either have or have not been lost. I explore the validity of this assumption, and highlight difficulties in using such experiments to draw conclusions about the ecosystem consequences of biodiversity loss in natural systems. Notably, these experiments do not mimic what happens in real ecosystems either when local extinctions occur or when species losses are offset by gains of new species. Despite limitations, this single experimental approach for studying how biodiversity loss affects ecosystems has often been advocated and implemented at the expense of other approaches; this limits understanding of how natural ecosystems respond to biodiversity loss. I conclude that a broader spectrum of approaches, and more explicit consideration of how species losses and gains operate in concert to influence ecosystems, will help progress this field.