There is growing evidence that the increase of wild ungulates worldwide is responsible of strong deleterious impacts on various component of forest biodiversity (tree and shrub regeneration, insects, songbirds). Plant communities are particularly affected by overbrowsing, and overabundant deer are increasingly recognized as a major factor of decline in understory plant diversity in temperate and boreal forest ecosystems. In grassland ecosystems, refuges from herbivores have been shown to provide suitable habitat for plants sensitive to browsing and grazing, and to allow the conservation of rare plant species. In the present study we analyze if such refuges can maintain representative samples of forest plant communities exposed to strong browsing pressure by deer. For this we used the unique situation offered by the presence, side by side, on the Haida Gwaii archipelago of forested islands colonized or not by introduced deer. We compared plant assemblages on rocky outcrop to assemblages found in the adjacent field-layer in order (1) to assess, on islands without deer, the potential of refuges to be representative of the local field-layer communities and (2), on islands colonized by deer for over 60 years, to analyze how far plant assemblages on rocky outcrops are representative of un-browsed field-layer communities and can serve as bioassays for field-layer plant communities in the absence of unbrowsed reference sites. We did this for forest-interior field-layer plant communities as well as for the forest edge plant communities found along the shorelines. We used refuges on treefall mounds to assess the role refuges can play as sources of propagules in two understory shrubs and ferns. Our results indicate that, in presence of deer, plant species density and plant cover are dramatically higher in refuges than in the adjacent field-layer plots. The comparison of these refuge communities to the reference state provided by the field-layer on islands without deer indicated that refuges maintain a representative part of the unbrowsed plant community, although with differences in the cover percentage in some plants. Some typical plants are more abundant in the refuges and, more sobering, six field-layer species are only poorly present in the rocky habitat. Despite these differences and their small size, refuges shelter a representative part of the understory and forest edge vegetation of the studied sites and could serve as a bioassay of deer overbrowsing in situations lacking a reference state.