1. The spatial structure of natural populations may profoundly influence their dynamics. Depending on the frequency of movements among local populations and the consequent balance between local and regional population processes, earlier work has attempted to classify metapopulations into clear-cut categories, ranging from patchy populations to sets of remnant populations. In an alternative, dichotomous scheme, local populations have been classified as self-sustaining populations generating a surplus of individuals (sources) and those depending on immigration for persistence (sinks). 2. In this paper, we describe the spatial population structure of the leaf-mining moth Tischeria ekebladella, a specialist herbivore of the pedunculate oak Quercus robur. We relate moth dispersal to the distribution of oaks on Wattkast, a small island (5 km(2)) off the south-western coast of Finland. 3. We build a spatially realistic metapopulation model derived from assumptions concerning the behaviour of individual moths, and show that the model is able to explain part of the variation in observed patterns of occurrence and colonization. 4. While the species was always present on large trees, a considerable proportion of the local populations associated with small oaks showed extinction-recolonization dynamics. The vast majority of moth individuals occur on large trees. 5. According to model predictions, the dominance of local vs. regional processes in tree-specific moth dynamics varies drastically across the landscape. Most local populations may be defined broadly as 'sinks', as model simulations suggest that in the absence of immigration, only the largest oaks will sustain viable moth populations. Large trees in areas of high oak density will contribute most to the overall persistence of the metapopulation by acting as sources of moths colonizing other trees. 6. No single 'metapopulation type' will suffice to describe the oak-moth system. Instead, our study supports the notion that real populations are often a mix of earlier identified categories. The level to which local populations may persist after landscape modification will vary across the landscape, and sweeping classifications of metapopulations into single categories will contribute little to understanding how individual local populations contribute to the overall persistence of the system.