To minimize the impacts of introduced pests and to justify and prioritize pest control, managers need to know the relationship between pest density and damage. This relationship can be difficult to quantify because pest impacts can be highly variable. In New Zealand, introduced brushtail possums (Trichosurus vulpecula) browse a wide range of native forest species. However, possum browse is extremely patchy making it difficult to quantify the relationship between density and damage, meaning the benefits of reducing possum densities are poorly understood. We quantified patterns of possum browse on kamahi (Weinmannia racemosa), a common canopy tree species, at 21 forest sites that were repeat-measured over an 8-year period in the North Island, New Zealand, during which time possum densities fluctuated widely. We fitted a multilevel statistical model in order to quantify the relationship between possum density and browse damage while simultaneously quantifying how browse varied among trees, sites and years. Higher possum densities were associated with greater browse damage, but browse was also patchily distributed among trees at the same site, and among sites and years for a given possum density. This heterogeneity meant there was no simple density damage relationship, with the relationship differing from tree to tree and among sites and years. Our results show that at most sites reductions in possum density would have little benefit in reducing the probability of heavy browse on kamahi trees, but at a few sites there would be substantial benefits.This approach provides insights into the pattern and potential causes of variability in possum impacts, and a quantitative basis for prioritizing sites for possum control.