Holton (2017) argues from the observation that no natural language has a contrafactive to the claim that the object of knowledge is a fact. His argument relies on the idea that if the truth inference of 'know' arises because its declarative complement denotes a fact, then the falsity inference of contrafactives would be due to their declarative complements denoting contrafacts. But, Holton thinks, there are no contrafacts. Thus, given Holton's claim about knowledge, there are no contrafactives. In reply, Hyman (2017) and Roberts and Özyildiz (2023) insist there are contrafacts. My aim is to argue that their reply is ineffective. Even if there are contrafacts, they do not get you contrafactives. On Holton's picture, natural languages only get contrafactives if they let some of their declarative complements denote contrafacts. But, I argue, there are good reasons why natural languages do not do this. I show that if a natural language lets some of its declarative complements denote contrafacts, it invites a systematic and particularly problematic form of ambiguity in factive attitude ascriptions. For instance, a knowledge ascription 'S knows p' could now attribute either knowledge of p or knowledge of not-p, and in many contexts that ambiguity would be irresolvable.