2005. Feeding patterns by roe deer and rabbits on pine, willow and birch in relation to spatial arrangement. Á/ Oikos 109: 513 Á/520.The location of a plant and its association with the surrounding vegetation may be a strong determinant of herbivore foraging decisions. The attractiveness of a food plant may be reduced if associated with a less preferred one (repellent Á/plant hypothesis) or, conversely, it may escape herbivory via association with a more preferred species (attractant Á/decoy hypothesis). In this study we tested the hypothesis that selection of the same food plants by two herbivore species with different body mass, i.e. roe deer and rabbits, is affected by the spatial disposition of preferred plants in relation to less preferred ones. We planted willows, birches and pines, representing food species of higher, intermediate and lower preference, respectively, in different spatial arrangements, to manipulate patch quality and accessibility to herbivores. Contrasting patches were constructed by planting willows or pines in the midst of birches and single-species arrays of all three species, in a blocked design, replicated six times in an area occupied by both roe deer and rabbits. Preference patterns were studied by recording browsing on current years' shoots and older plant parts. Across all trees in the experiment, we confirmed that roe deer clearly preferred willow over birch, pine was not browsed at all. There was greater herbivory by roe deer on birches occurring peripherally in an array, when the associated central species was willow or pine, as compared to birches in a single-species birch patch. Therefore food choices by roe deer appear to be determined by both patch and individual level selection. The rabbits only showed preference at a species level and did not respond to spatial arrangement of trees. The results from this study indicate that the location and association of trees can strongly affect foraging patterns of roe deer, but not in a manner wholly consistent with the repellent Á/plant or the attractant Á/decoy hypotheses.