The stress-gradient hypothesis predicts that the intensity of interspecific positive interactions increases along environmental severity (i.e. stress and disturbance) gradients faster than the intensity of negative interactions. This study is the first to test if the stress-gradient hypothesis is supported for a location in the climatically extreme and species-poor sub-Antarctic. To do so, we investigate the fine-scale spatial distribution of plant species across altitude- and aspect-related abiotic severity gradients on a scoria cone on Marion Island. A clear altitudinal severity gradient was observed across the scoria cone, with lower temperatures, stronger winds and greater soil movement at higher altitudes. The altitudinal severity gradient was matched by stronger interspecific spatial association between the four dominant species at higher altitudes and in areas of lower vegetation cover. This suggests that, relative to the intensity of competition, the intensity of facilitation is greater under more severe conditions, supporting the stress-gradient hypothesis at the community level (i.e. for multiple pairs of species) and corroborating its usefulness for predicting variation in plant interactions at high latitudes and altitudes. Furthermore, the directional intraspecific aggregation and interspecific association plant cover patterns found within the gradient suggest that protection from the prevailing wind and from burial by loose substrate are the dominant facilitative mechanisms. Thus, plants benefit from the presence of neighbours when they provide shelter and substrate stability, and the relative intensity of this positive interaction is greatest at higher altitudes, and varies between species pairs. This study, therefore, not only provides support for the stress-gradient hypothesis in the sub-Antarctic, but also demonstrates fine-scale directional spatial patterns between multiple species nested within the severity gradient.