Conversational implicatures are easy to grasp for the most part. But it is another matter to give a rational reconstruction of how they are grasped. We argue that Grice's attempt to do this fails. We distinguish two sorts of cases: (1) those in which we grasp the implicature by asking ourselves what would the speaker have to believe given that what he saidis such as is required by the talk exchange; (2) those in which we grasp the implicature by asking ourselves why it is that what the speaker said is so obviouslynot such as is required by the talk exchange. We argue that Grice's account does not fit those cases falling under (2).