In his attack on neo-Kripkean accounts, McDowell has accepted that attributions of intention are normative, in the same sense in which attributions of meaning are normative, in the same sense in which attributions of meaning are normative. I will argue that this is a wrong assimilation. By referring to certain of Wittgenstein's ideas on intentionality (circa 1930) that were preserved in Philosophical Investigations, I will try to track an argument from which it follows that expressive behaviour is the proto-phenomenon of intentionality. The features of this notion justify McDowell's ideas about the impossibility of grounding the 'bedrock' of grammatical conventions. Nevertheless, the underlying reasons for such impossibility are slightly different from those that McDowell has defended