Social structure can have profound evolutionary and ecological implications for animal populations. Structure can arise and be maintained via social preferences or be indirectly shaped by habitat structure. Quantifying the drivers of social structure is important to understand how social networks can shape evolutionary landscapes. Here, we study a large community of wild birds fitted with uniquely-coded passive integrated transponder (PIT) tags and recorded on a grid of automated feeders fitted with radio frequency identification (RFID) antennae. These data reveal that preferred movement pathways between sites that are consistent between years and not predicted by habitat features alone drive between-year consistent multi-level community structure in the social network. Our study highlights how ecological factors can shape social structure at the population scale, which has widespread implications for understanding eco-evolutionary dynamics.