Marr's celebrated contribution to cognitive science (Marr 1982, chap. 1) was the introduction of (at least) three levels of description/explanation. However, most contemporary research has relegated the distinction between levels to a rather dispensable remark. Ignoring such an important contribution comes at a price, or so we shall argue. In the present paper, first we review Marr's main points and motivations regarding levels of explanation. Second, we examine two cases in which the distinction between levels has been neglected when considering the structure of mental representations: Cummins et al.'s distinction between structural representation and encodings ) and Fodor's account of iconic representation (Fodor 2008). These two cases illustrate the kind of problems in which researchers can find themselves if they overlook distinctions between levels and how easily these problems can be solved when levels are carefully examined. The analysis of these cases allows us to conclude that researchers in the cognitive sciences are well advised to avoid risks of confusion by respecting Marr's old lesson.Ever since Marr's influential work on the computational account of vision (Marr 1982) it has been a familiar idea that computational research can be taken to involve (at least) three levels of description/explanation. These are: the level of the function to be computed; the level of the algorithm that computes the given function; and finally, the level of realization of the function in hardware structures.

No abstract

No abstract

No abstract