In two experiments, it is demonstrated that knowledge about idiosyncratic aspects of one's own memory performance may become the basis for inferences in recognition. In the first study, beliefs about the effect of the encoding conditions on memory were experimentally induced by varying the memory task such that participants' performance was superior either for high-frequent or for low-frequent words. As a consequence, participants tended to accept test words that belonged to the category that they were led to believe to be less memorable. In the second study, idiosyncratically based beliefs were compared to nomothetically based beliefs and it was shown that idiosyncratic knowledge had a predominant influence. These sets of findings suggest that to understand reconstructive mechanisms in recognition, idiosyncratic knowledge (or beliefs) about one's own memory must be taken into account.