Abstract Questions: 1. Do harvester ants (Messor barbarus) promote seed mortality in Mediterranean grassland?; 2. Is this effect greater in large-seeded species? Location: Central Spain. Methods: We established an ant-exclusion experiment of five circular (1.5 m diameter) plots from where ants were excluded during one year, along with ten control plots. We recorded the seed bank of all species in the plots both before and after the treatment. The effect of seed length and weight was analysed after transforming data into phylogenetically independent contrasts, and alternatively by dividing the species data set into morphological groups. Results: Longer and heavier seeded species significantly increased in the seed banks under the exclusion treatment, although ants did not significantly modify overall seed densities. Conclusions: Although the ants do not collect large numbers of seeds, they differentially affect the composition of the seed banks by selecting the longest or heaviest seeds, or both. The persistence of this short-term effect in the seed bank may result, over a number of years, in the system evolving towards a predominance of small-seeded annuals, congruent with the species composition actually observed in Mediterranean grasslands. Abbreviations: PIC = Phylogenetically independent contrast. Nomenclature: Castroviejo (1986–2005), except in the taxa yet to be covered which are named following Tutin et al. (1964–1980).