A supporter of a candidate for office may wish to restrict her power or to send her a message by casting a protest vote against her. A sufficiently large protest may convince the candidate to accept the protesters' demands af- ter winning the election; however, if the protest is too large then it risks causing the candidate to lose to a disliked opponent. I study a model of protest voting in which there is uncertainty about the true electorate-wide enthusiasm for the protest relative to the conflicting desire to ensure that the candidate is safely elected. I find that protest votes are strategic substitutes, and that protest vot- ing reacts negatively to voters' expectations about the true enthusiasm for the protest. An increase in the candidate's popularity (and so a reduction in the desire for a successful protest relative to the wish to see her elected) is offset by increased protest voting. Greater popularity can increase protest voting by enough to harm a candidate's performance at the ballot box. 1. THE LOGIC OF PROTEST VOTING In a two-horse-race election, a voter's incentives seem straightforward: if he wishes his fa- vorite to win then he should vote for her. Nevertheless, sometimes voters support the leading opponent, they vote for an out-of-the-running third candidate, or they spoil their ballot pa- pers. In this paper I use a theoretical model to study how these acts of protest voting respond to the electoral environment, to beliefs about the candidate's popularity, and to voters' antici- pation of the candidate's reaction to the election result. Amongst other results, I find that any gain in a candidate's popularity relative to the popular enthusiasm for a successful protest is offset by an endogenous increase in protest voting. Furthermore, if the candidate responds endogenously to the election (by this I mean that she infers her popularity from the outcome, and then decides whether to give in to the protest's demands) then, perhaps surprisingly, the increase in protest voting is large enough to result in a net fall in her ballot-box performance.