This paper studies voting behavior in proportional representation systems. It proposes a model in which coalition formation cannot be accurately predicted once the distribution of seats in parliament is known. Voters care about policy outcomes, and are all assumed to be strategic. Identical agents who expect extremely similar electoral outcomes but expect different coalitions have radically opposed strategies. The model is tested against empirical data from Israel, 2006. Results suggest that agents with similar preferences who have reasonably similar expectations regarding the distribution of votes may use different voting strategies precisely when expect different coalitions. The implications of this are discussed.