Resolving links between subsequent referents (e.g., the car) and open discourse roles (as in Keith drove to London yesterday. The car kept overheating) is crucial for discourse understanding. This article investigates the contribution of lexical semantic factors (e.g., that drive implies using a vehicle) as compared to more general contextual factors in the on-line resolution of such links. We report an eye-tracking experiment that measures immediate and delayed effects of both kinds of information as readers resolve the reference. The results indicate that lexical information dominates the initial linking process with more general contextual influences emerging later. They are discussed in terms of the distinction between early bonding and subsequent resolution processes that has been proposed for other kinds of anaphoric interpretation (Sanford, Garrod, Lucas, & Henderson, 1983).