Large herbivores and their grazing impact is an important subject in ecology. It is well known that grazing shapes the structure, diversity and functioning of terrestrial ecosystems, but how the level of density dependence on the herbivore might change due to the interac-tion between herbivore and habitat is less studied. In this thesis, I studied feedback process-es in an alpine environment in Southern Norway. In a large scale, long term experimental setting, sheep have been kept at high and low grazing densities for ten consecutive grazing seasons. I investigated how density-dependent habitat selection by a large grazing herbivore, sheep (Ovis aries), develops over the time scale of a decade. By applying biomass cages I could study how contrasting grazing levels influence biomass production. I predicted that sheep habitat selection would be density dependent, affected by annual and seasonal varia-tions, and show a trend in density dependent selection for given patches within the same vegetation type. With respect to effects on biomass, I predicted that biomass production would be highest at low grazing density for tolerant functional groups, and that resistant functional groups would decrease with increasing density. These predictions were investigat-ed using generalized linear mixed effect models. I found that habitat selection was density dependent. Seasonal and annual variations had a significant impact on sheep habitat selec-tion. Graminoid and herb biomass significantly changed with increasing sheep density. There was a trend in change in long-term within-patch use, but sheep habitat selection was more affected by seasonal and annual climatic variations. This study sheds light on the little studied interactive effects of herbivores and habitat. My findings increase the general com-prehension of density dependent habitat selection and how this affects biomass production, and helps us understand how the habitat selection by herbivores is affecting the ecosystem productivity depending on the population density over longer time frames.