Formal analysis predicts that the likelihood of an electoral accident depends on the preference intensity for a successful protest, but not on the protest's popularity: an increase in protest's popularity is fully offset by a reduction in the individual probability of casting a protest vote. By conducting the first laboratory experiment on protest voting, we find strong evidence in favor of the first prediction and qualified support for the latter. While the offset effect is present, it is not as strong as the theory predicts: protest candidates gain both by fanaticising existing protesters
and by expanding the protest's popular base.