Interspecific interactions can vary within and among populations and geographic locations, and this variation can influence the nature of the interaction (e.g. mutualistic vs. antagonistic) and its evolutionary stability. Globeflowers are exclusively pollinated by flies, whose larvae feed only on their seeds. Here we document geographic variability in costs and benefits in globeflowers in sustaining their pollinating flies throughout the range of this arctic-alpine European plant over several years. A total of 1,710 flower heads from 38 populations were analysed for their carpel, egg and seed contents. Individual and population analyses control for the confounding influences of variation in both: (1) population traits, such as fly density and egg distribution among flower heads; and (2) individuals traits, such as carpel and egg numbers per flower head. Despite considerable variation in ecological conditions and pollinator densities across populations, large proportions (range 33-58%) of seeds were released after predation, with a benefit-to-cost ratio of 3, indicating that the mutualism is stable over the whole globeflower geographical range. The stability of the mutualistic interaction relies on density-dependent competition among larvae co-developing in a flower head. This competition is revealed by a sharp decrease in the number of seeds eaten per larva with increasing larval number, and is intensified by non-uniform egg distribution among globeflowers within a population. Carpel number is highly variable across globeflowers (range 10-69), and flies lay more eggs in large flowers. Most plants within a population contribute to the rearing of pollinators, but the costs are greater for some than for others. Large globeflowers lose more seed to pollinator larvae, but also release more seed than smaller plants. The apparent alignment of interests between fly and plants (positive relationship between numbers of seed released and destroyed) is shown to hide a conflict of interest found when flower size is controlled for.