We investigated whether memory for scientific arguments and their sources were affected by the appropriateness of the claim-evidence relationship. Undergraduates read health articles in one of four conditions derived by crossing claim type (causal with definite qualifier, associative with tentative qualifier) and evidence type (experimental, correlational). This manipulation produced articles that overstated the results of a study and articles that understated their results, along with appropriate controls for each. We found that evidence and, to a lesser extent, source information was recalled more poorly for articles that overstated results (i.e., causal claims using correlational evidence) than for those where evidence was appropriate (i.e., causal claim with experimental evidence). Readers rejected these overstatements based on the study design rather than reprocessing the text. In contrast, understatements (tentative claim, experimental evidence) were recalled just as well as their appropriate control; however, the target content was reprocessed at a higher rate. These findings suggest that readers may remember an inappropriate definite conclusion but fail to recall the evidence used to support it.