Although polling accuracy increases throughout the election, polls are always at least a little wrong on election day. In this article, we attempt to understand how characteristics of particular elections may make them harder (or easier) to predict. In particular, we focus on estimating the impact of voter turnout, electoral change, and vote buying on polling error. We find support for two of the three hypotheses. There is little evidence that voter turnout affects polling error. However, polling errors tend to be higher where there have been large changes in parties’ vote share from the previous election. We also find that higher prevalence of vote buying may be associated with larger polling errors.