The main objective of this study was to address questions with regard to how the soil seed banks and vegetation change along a habitat gradient. In this article, we chose the Thermopsis lanceolata community in degraded alpine arid grasslands as our research subject and determined the relationship between the alpine arid grassland soil seed bank and above-ground vegetation in the Qilian Mountains using field investigation and laboratory analysis. Fifteen species were present in the above-ground vegetation and 18 species were present in the soil seed bank. Dominant above-ground vegetation species included edible grasses such as Stipa krylovii, Aneurolepidium dasystachys and Agropyron cristatum, and non-edible grasses such as Thermopsis Lanceolata and Heteropappus altaicus; however, no other species was observed to exhibit any distinct dominance. Moreover, the degree of similarity between the above-ground vegetation and the soil seed bank was distinctive. The Sorenson similarity index was low between the above-ground vegetation and the soil seed bank along the vegetation gradient. The Sorenson similarity index between the soil seed bank and observed vegetation functional groups exhibited varying characteristics. The number of edible grasses was greater than that of non-edible grasses in the soil seed bank (except V gradient). The number of species of vegetation was also greater than that of non-edible grasses (except Ⅲ gradient). The number of species in the soil seed bank was not lower than that of the above-ground vegetation species. S. krylovii and A. dasystachys were founded in the above-ground vegetation but not in the soil seed bank, whereas Poa annua, Chenopodium glaucum, Allium polyrhizum, Rumex patientia, Lepidium apetalum, and Gentiana macrophylla were found in the soil seed bank but not in the vegetation. The proportion of biomass attributable to gramineous plants was low for edible grasses, whereas the biomass of poisonous weeds was significantly greater in the vegetation gradient than in the soil seed bank. The plant population biomass had no unanimous relation with seed amount. Edible grass and non-edible grass were significantly negatively correlated (r=-0.940,P=0.018) along the vegetation gradient, but changes in the soil seed bank density were not correlated. The relationship between the biomass of edible grass and the soil seed bank density was uncorrelated. The biomass of non-edible grass and the soil seed bank density were significantly correlated along the vegetation gradient (r=0.902,P=0.036). Our research indicated that the seed bank could retain and restore several species in the process of grassland degradation, and these species could be valuable for vegetation succession and grassland restoration.