In this paper we experimentally investigate the consequences of electoral fraud on voter turnout. The experiment is based on a strategic binary voting model where voters decide whether to cast a costly vote in favour of their preferred candidate or to abstain. Minority candidate can illicitly influence the electoral process by applying ballot-box stuffing. In the experiment we implement two different framings: we compare voter turnout in a neutral environment and with framed instructions to explicitly replicate elections. This approach enables to both test the model's predictions and to estimate framing effects of voting and fraud. Comparison of experimental results with theoretical predictions reveals over-voting, which is exacerbated when fraud occurs. Moreover, turnout increases with moderate level of fraud. However, when considering higher electoral fraud, theoretical predictions are not matched. Voters fail to recognize that the existence of a relatively larger number of "agents" voting with certainty considerably decreases the benefits of voting.Importantly, framing matters, as revealed by the higher turnout of those in the majority group, against which the fraud is applied. Finally, individual level regression analysis provides evidences of strategic voting.