Meadow microcosms were established from seed on low-fertility soil of known seed bank composition, and subjected to manipulations of simulated grazing, cutting date, temperature and fertility for seven years. The composition and density of the seed bank was then determined in five 2-cm soil layers (0–2, 2–4, 4–6, 6–8 and 8–10 cm). The seed bank contained three distinct groups of species: species present in the original soil, sown species, and ‘others’. The seed bank was little affected by the experimental treatments, presumably because the sown species made only a small contribution to the seed bank. Nearly all the species in the original soil are known to possess persistent seed banks and had survived, although at reduced density, for seven years. Density of the most abundant species in this group, Sagina procumbens, had changed very little over seven years, confirming the well-documented longevity of the seeds of this species. Seeds of sown species made up only about a quarter of the seed bank, despite accounting for virtually all the above-ground vegetation. Of the sown meadow species, only Plantago lanceolata and Alopecurus pratensis were relatively abundant in the seed bank. These results strongly support the conclusion of other authors that most meadow species, once lost owing to the effects of fertilizers or inappropriate management, will not reestablish from the seed bank. Among species which were neither sown nor present in the original soil, the majority possessed adaptations for wind dispersal and had presumably dispersed into the experimental plots from outside. The most abundant member of this group, Betula pendula, had dispersed from a nearby tree. Density of Betula seeds declined sharply with depth, consistent with the view that seeds on the soil surface are rapidly lost, mainly through germination, but seeds that become buried survive much better. Seeds of Betula appear to be persistent but not particularly long-lived.