When directed to ignore evidence of a witness's previous bad character because of a violation of the rules of evidence, are jurors' beliefs still affected? The intuition is that they will be because in everyday argumentation, fallacies, like the ad hominem, are effective argumentative strategies. An ad hominem argument (against the person) undermines a conclusion by questioning the character of the proposer. This intuition divides current theories of argumentation. According to pragmadialectical theory (e.g., Van Eemeren & Grootendorst, 2004), procedural rules exactly like the rules of evidence are part of our cognitive resources for evaluating arguments. If one of these rules is violated, an argument should be treated as a fallacy and so it should not alter someone's belief in the conclusion. Some recent experiments investigating how reasonable these arguments are perceived to be seem to support this account (van Eemeren, Garssen, & Meuffels, 2009). These experiments are critiqued from the perspective of the relevance (Walton, 2009, 2010) and epistemic (Hahn & Oaksford, 2006, 2007; Oaksford & Hahn, 2004) approaches to argumentation. An experiment investigates the predictions of these approaches for a graded belief change version of van Eemeren et al.'s (2009) experiment, and the results are modeled using a Bayesian congruent prior model. These results cannot be explained by the pragmadialectical approach and show that in everyday argument people are extremely sensitive to the epistemic relevance of evidence. Moreover, it seems highly unlikely that this can be switched off in more formal contexts such as the courtroom.