Predicting spatial distribution of large herbivore foraging is important for successfulmanagement, but accurate predictions remain elusive against a background of multiple causes modifiedby environmental stochasticity. Moose (Alces alces) might prefer to browse areas with high plant density,but if snow depth co-varies with plant density, this could restrict access to these sites and force useof sites with lower plant density and snow depth. Moose browsing was measured in 72 plots distributedwithin the subarctic birch (Betula spp.) forest landscape at Abisko in northern Sweden in 1996. In 2010,the same plots were revisited and the measurements repeated. A generalized linear model predictedmoose browsing on birch in 2010 from the browsing pattern on birch measured in 1996. The modelsuggested that neither total density of willow and birch stems nor snow depth were influential of foragingdistribution of birch at multiple spatial scales. The spatial scale at which clustering of browsing on birchoccurred, coincided with the scale of clustering of birch and willow (Salix spp.) stems at distances of1000–2500 m; at lesser distance browsing was distributed randomly. We concluded that moose demonstratestability in spatial browsing patterns after 14 years which corresponds to 3–4 generations of moose,and that plant density represents a cue for moose only at certain scales. Predictability of feeding sites isvaluable for long-term moose and forest management, and conservation planning.