The aim of this paper is to show that some current theoretical frames for the causal explanation of rational action, that is, the accounts of Davidson and Fodor, cannot be successful in their attempts to answer satisfactorily the following question: "Why does our strategy in explaining rational action by ascribing beliefs and desires works?". The reason of the failure, in my opinion, could be that one: the proposals mentioned above try to account simultaneously for the causal and rational features of human action. Taking a different path, I propose, with D. Dennett, that there is room in cognitive science for two different projects, each one accounting in its own way for the rational and causal features separately. Moreover, if we want to preserve a realist conception of representations, a three level (at least) cognitive architecture must be posited, being the two lower levels of connectionist sort, and classical the highest one, however modified in its ideal assumptions.