On the face of it, conflicting constraints are placed on agents' knowledge of their own action: it is demanded that that which is known is "on the outside", but that the way in which it is known is "from the inside". I propose to look at the way in which Anscombe sets up this epistemological paradox and attempts to solve it. I discuss two ways in which Anscombe proposes to dissolve the paradox of agents' knowledge, whereof the first one is rejected. Finally, I discuss different problems for the second way and suggest that we can save the Anscombian framework by rethinking the role of perception in action.