Many theories of the de dicto / de re ambiguity for quantifiers and descriptions follow the tradition started by Kaplan and L Lewis in that they make use of notions that are epistemic in nature, such as the notion of acquaintance. This may create the impression that the question about de re in attitude report semantics should always be resolved by looking at the reported attitude; if the latter qualifies as de re according to some epistemological criteria, then also the attitude report may be true de re. The present paper aims to provide an argument to the contrary. The argument proceeds in several steps. First, we point out the existence of a wide range of de re readings (mostly already known from the literature), some of which do not target referential or quantificational expressions at all. Second, we show that the existing analyses either give wrong predictions for such cases, or are incomplete, or at least inelegant. Third, we offer a new analysis (which, as it turns out, has predecessors not united by any single tradition) whose main ingredient is the observation that the right predictions result from modifying the Context Set of the conversation in certain intuitive ways before the attitude report is added to the Common Ground. This is the semantic contribution of the paper. The philosophical upshot is that the reasons for an attitude report being de re at least in some cases have little to do with the grounds on which the corresponding mental state - the attitude itself - qualifies as de re. We also speculate as to how the proposed analysis, which divorces de re'ness from epistemic rapport, may be extended onto more traditional cases.