Abstract The “habitat‐specific species pool hypothesis” proposes that differences between habitats in the sizes of their species pools are the main drivers of diversity responses to habitat heterogeneity. Empirical tests of this hypothesis are not trivial as species might be missing from ecologically suitable habitats due to limited dispersal, while others may occur in unsuitable habitats by means of source–sink dynamics and mass effect. We tested the habitat‐specific species pool hypothesis in a local, environmentally heterogeneous community of annual plants using a novel “ecological selection” experiment. Mixtures of seeds representing the whole community were sown in each habitat, and the emerging species were exposed to six generations of selection by environmental filtering and competition while being blocked from dispersal. A comparison of the total number of species that were able to survive in each habitat (i.e. to pass the selection test) with data on species richness in the natural community allowed us to test the degree to which observed differences in species richness between habitats could be explained by differences in the sizes of the respective species pools. Results supported the species pool hypothesis, showing that differences in the sizes of the habitat‐specific species pools were important in determining diversity responses to habitat heterogeneity. Moreover, species richness showed a unimodal response to local‐scale gradients in habitat productivity, and this response could be attributed to underlying differences in species pool sizes. Both results were robust to properties of the data and the method of analysis. Synthesis. Our results provide a strong experimental evidence that differences in the sizes of habitat‐specific species pools might be important in shaping the diversity of local communities. Future theoretical and empirical studies in community ecology should explore the potential sources and implications of such differences.