Women are under-represented in politics. In this paper, we test one of the potential explanations for this situation: gender-bias from voters. We use a natural experiment during French local elections in 2015: for the first time in this country, candidates had to run by pairs, which had to be gender-balanced. We argue that this reform confused some voters, who might have assumed that the first name on the ballot represented the "main" candidate. Since the order of the candidates on the ballot was determined by alphabetical order, the order of appearance of male and female candidates is as-good-as-random, and this setting allows us to isolate gender biases from selection effects. Our main result is that there exists a negative gender bias affecting right-wing candidates, who receive about 1.5 points lower shares of vote when the female candidate appears first on the ballot. The missing votes prevented some pairs of candidates from being elected. Using the fact that the candidates can (but do not have to) report additional information about themselves on the ballot, we show that this discrimination is likely to be statistical, since the most affected women are those running in pairs which do not report any information.