Abstract. The seed content of soils is often estimated through germination tests, though these methods are always somewhat inaccurate due to the presence of dormant seeds in the samples. The researcher thus faces the question of whether to continue the germination test or to stop it in the search for an accuracy-to-effort balance. In this paper I analyze the accuracy of seed content estimates obtained after a first-year germination test, by comparison to the germination recorded after three-year cultivation, in 48 soil seed bank samples and 389 from herbivore dung. After the first 9-month cultivation, I recorded 85 ± 1% seedlings and 90 ± 1% species in soil samples, while the accuracy in those of dung was significantly lower, 48 ± 1% seedlings and 65 ± 1% species. The accuracy of estimations varied among samples within experiments, with significant differences in the estimation of species richness in both cases. I did not find consistent differences in the accuracy of estimations linked to seedling densities in growing pots, but the taxonomic composition of samples was a major source of bias. Thus, 22% and 36% of the most frequent species showed germinabilities in the first year significantly different from the rest, and some generalities arose, like the high germinability of grasses and the hardseededness of legumes. I would thus recommend the use of at least two germination cycles for seed bank estimations and a cautious approach when comparing samples with very different origin and/or taxonomic composition.