Geographical gradients of patterns of species associations in ecological communities are largely unknown. Previous evidence indicated nested community assembly — caused mainly by unequal colonization probabilities and habitat capacity — and a tendency towards negative species associations in arid and tropical plant and animal communities. Patterns of community assembly in arctic environments are poorly studied. Here we use a data set on arctic plant and animal species obtained from arctic islands of the Kandalaksha Bay (White Sea), to infer patterns of species association across taxa and trophic groups. We performed co-occurrence and nestedness analyses to study patterns of community assembly and diversity of 1109 plant and animal species grouped according to taxa, dispersal ability, and ecological guild membership. Twelve out of 50 (24%) sufficiently species-rich families and orders on the environmentally relatively stable forested islands showed significantly negative species associations (segregation), while this proportion decreased to less than 13% on less stable heath, rocky, and sea-shore islands. Segregation was not linked to spatial species turnover across islands. Species richness of plants and animals decreased at higher levels of disturbance. We detected evidence for a gradient in species richness and ecological interactions from the most disturbed sea-shore and rocky islands to more stable forested islands. Species spatial distributions appeared to be largely random, in contrast to previous meta-analyses that used mainly communities at lower latitudes. We speculate that in arctic environments spatial turnover of species (vicariant segregation) is of less importance than turnover-independent (checkerboard) segregation. Our data support the view that ecological assemblages in high-latitude environments are less structured by ecological interactions than comparable assemblages in lower latitudes. We also add to the evidence that environmental disturbance regimes work against stable community structures. We notice the need for a formal meta-analysis on latitudinal trends in community structure.