The dynamics of seed population on slope surfaces were studied to test the hypothesis that the lack of vegetation on badlands is caused by seed removal by erosion. The initial soil seed bank and two years of seed rain and seed removal by erosion were estimated in two small catchments, and a seed balance was constructed. In addition, six rainfall simulation experiments were performed to test the susceptibility of seeds to be removed by overland flow. A variety of soil surface conditions, rainfall characteristics and plot sizes were used in these experiments. Soil seed bank densities are low, but enough for the development of plant cover. Seed losses due to erosion after natural rains were low (< 13 %), and in agreement with seed losses from simulated rainfall experiments. After two years, seed inputs in the seed rain were greater than seed outputs through seed removal, which resulted in a continuous increase in the numbers of seeds in the soil bank. These results point out that seed removal by erosion is not the key factor explaining the lack of vegetation on badlands. It is suggested that other factors, such as those related to seed germination and seedling survival, may play an important role.