Flocks of birds in flight represent a striking example of collective behaviour. Models of self-organization suggest that repeated interactions among individuals following simple rules can generate the complex patterns and coordinated movements exhibited by flocks. However, such models often assume that individuals are identical and interchangeable, and fail to account for individual differences and social relationships among group members. Here, we show that heterogeneity resulting from species differences and social structure can affect flock spatial dynamics. Using high-resolution photographs of mixed flocks of jackdaws, Corvus monedula, and rooks, Corvus frugilegus, we show that birds preferentially associated with conspecifics and that, like high-ranking members of single-species groups, the larger and more socially dominant rooks positioned themselves near the leading edge of flocks. Neighbouring birds showed closer directional alignment if they were of the same species, and neighbouring jackdaws in particular flew very close to one another. Moreover, birds of both species often flew especially close to a single same-species neighbour, probably reflecting the monogamous pair bonds that characterize these corvid social systems. Together, our findings demonstrate that the characteristics of individuals and their social systems are likely to result in preferential associations that critically influence flock structure. 2013 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved. How do large aggregations of individuals, each of which may differ in its preferred outcome, coordinate their movements? The spectacular displays of flocking birds led the naturalist Edmund Selous (1931) to postulate a role for ‘thought transference’, but recent advances have begun to unravel the mysteries of collective movement without appealing to the supernatural (Couzin & Krause 2003; Conradt & Roper 2005; Sumpter 2006). Models of selforganizing systems suggest that repeated interactions among individuals following simple rules can generate complex patterns and coordinated group movements. Models of agents following simple rules of (1) long-range attraction to group members, (2) shortrange repulsion and (3) alignment between close neighbours have generated realistic representations of collective animal movements (reviewed in Sumpter 2006; Petit & Bon 2010). However, empirical verification of their assumptions remains scarce and largelyconfined to model systems such as starlings, Sturnus vulgaris