The initiation of translation in eukaryotes generally involves the recognition of a 'cap' structure at the 5' end of the mRNA. However, for some viral and cellular mRNAs, a cap-independent mechanism occurs through an mRNA structure known as the internal ribosome entry site (IRES). Here, I postulate that the first eukaryotic mRNAs were translated in a cap-independent, IRES-driven manner that was then superseded in evolution by the cap-dependent mechanism, rather than vice versa. This hypothesis is supported by the following observations: (i) IRES-dependent, but not cap-dependent, translation can take place in the absence of not only a cap, but also many initiation factors; (ii) eukaryotic initiation factor 4E (eIF4E) and eIF4G, molecules absolutely required for cap-dependent translation, are among the most recently evolved translation factors; and (iii) functional similarities suggest the evolution of IRESs from spliceosomal introns. Thus, the contemporary cellular IRESs might be relics of the past.