Scholars disagree about the nature of party attachments, viewing partisanship as either a social identity or a rational maximization of expected utility. Empirically, much of this debate centers on the degree of partisan stability: findings of partisan fluctuations are taken as evidence against the social-identity perspective. But drawing such conclusions assumes that the objects of identity—parties—are fixed. If we instead allow party brands to change over time, then partisan instability is consistent with a social-identity conception of partisanship. To demonstrate this, I develop a branding model of partisanship in which voters learn about party brands by observing party behavior over time and base their psychological attachment to a party on these brands. The model suggests that convergence by rival parties, making their brands less distinguishable, should weaken party attachments. I test this implication using a survey experiment in Argentina and find evidence consistent with the model.