We analysed the combined effects of pollination and seed predation on seed set of Centaurea jacea in 15 landscapes differing in structural complexity. In the centre of each landscape, a patch of Centaurea plants was established for standardized measurements of flower visitation, seed predation and seed set. Both the number of flower–visiting bees and the proportion of flower heads damaged by seed predators increased with landscape complexity, which was measured as the proportion of semi–natural habitats. The mean number of seeds per flower head did not increase with the proportion of semi–natural habitats, presumably because of the counterbalancing effects of pollination and seed predation. For a subset of undamaged flower heads, the number of seeds per flower head was positively correlated with the number of flower visits. Further reasons for the unexpected failure to detect a correlation between landscape complexity and seed set appeared to be changes in flower–visitor behaviour and the contrasting responses of honeybees and wild bees to habitat context. Landscape analyses at eight spatial scales (radius of landscape sectors, 250–3000 m) showed that different groups perceived the landscape at different spatial scales. Changes in pollinator numbers could be explained only at small scales (up to 1000 m), while the seed predators also responded to large scales (up to 2500 m).