Here is a widespread but controversial idea: those animals who represent correctly are likely to be selected over those who misrepresent. While various versions of this claim have been traditionally endorsed by the vast majority of philosophers of mind, recently, it has been argued that this is just plainly wrong. My aim in this paper is to argue for an intermediate position: that the correctness of some but not all representations is indeed selectively advantageous. It is selectively advantageous to have (...) correct representations that are directly involved in bringing about and guiding the organism's action. I start with the standard objection to the claim that it is selectively advantageous to represent correctly, the ‘better safe than sorry’ argument gand then generalize it with the help of Peter Godfrey Smith's distinction between Cartesian and Jamesian reliability and the trade‐off between them. This generalized argument rules out a positive answer to our question at least as far as the vast majority of our representational apparatus is concerned. But this argument fails to apply in the case of the correctness of a special kind of representation, one that is directly involved in bringing about, and in guiding, actions. We can still maintain that it is selectively advantageous to have a reliable mechanism for producing representations of this kind. (shrink)