Standard pairwise coreference resolution systems are subject to errors resulting from their performing anaphora identification as an implicit part of coreference resolution. In this paper, we propose an integer linear programming (ILP) formulation for coreference resolution which models anaphoricity and coreference as a joint task, such that each local model informs the other for the final assignments. This joint ILP formulation provides fscore improvements of 3.7-5.3% over a base coreference classifier on the ACE datasets.