In this paper I argue, against some recent criticisms, that the factivity of 'knows' does not support the traditional truth condition on knowledge. I articulate a conception of the factivity of 'knows' on which it is a matter of cancelable pragmatic presupposition, and consider the epistemological implications of this.