Of all the steps in mRNA translation, initiation is the one that differs most radically between prokaryotes and eukaryotes. Not only is there no equivalent of the prokaryotic Shine-Dalgarno rRNA-mRNA interaction, but also what requires only three initiation factor proteins (aggregate size approximately 125 kDa) in eubacteria needs at least 28 different polypeptides (aggregate >1600 kDa) in mammalian cells, which is actually larger than the size of the 40 S ribosomal subunit. Translation of the overwhelming majority of mammalian mRNAs occurs by a scanning mechanism, in which the 40 S ribosomal subunit, primed for initiation by the binding of several initiation factors including the eIF2 (eukaryotic initiation factor 2)-GTP-MettRNA(i) complex, is loaded on the mRNA immediately downstream of the 5'-cap, and then scans the RNA in the 5'-->3' direction. On recognition of (usually) the first AUG triplet via base-pairing with the Met-tRNA(i) anticodon, scanning ceases, triggering GTP hydrolysis and release of eIF2-GDP. Finally, ribosomal subunit joining and the release of the other initiation factors completes the initiation process. This sketchy outline conceals the fact that the exact mechanism of scanning and the precise roles of the initiation factors remain enigmatic. However, the factor requirements for initiation site selection on some viral IRESs (internal ribosome entry sites/segments) are simpler, and investigations into these IRES-dependent mechanisms (particularly picornavirus, hepatitis C virus and insect dicistrovirus IRESs) have significantly enhanced our understanding of the standard scanning mechanism. This article surveys the various alternative mechanisms of initiation site selection on mammalian (and other eukaryotic) cellular and viral mRNAs, starting from the simplest (in terms of initiation factor requirements) and working towards the most complex, which paradoxically happens to be the reverse order of their discovery.