Abstract. Five structural features in mRNAs have been found to contribute to the fidelity and efficiency of initiation by eukaryotic ribosomes. Scrutiny of vertebrate cDNA sequences in light of these criteria reveals a set of transcripts-encoding oncoproteins, growth factors, transcription factors, and other regulatory proteins-that seem designed to be translated poorly. Thus, throttling at the level of translation may be a critical component of gene regulation in vertebrates. An alternative interpretation is that some (perhaps many) cDNAs with encumbered 5' noncoding sequences represent mRNA precursors, which would imply extensive regulation at a posttranscriptional step that precedes translation .NITIATIO N oftranslation in multicellular eukaryotes is influenced by five aspects ofmRNA structure : (a) the m7G cap (355); (b) theprimary sequence or context surround ing the AUG codon (187,190,194); (c) the position of the AUG codon, i.e., whether it is the first AUG in the message (186); (d) leader length (198,199) ; and (e) secondary structure both upstream (188, 195) and downstream (196) from the AUG codon . Elsewhere (200) I have reviewed the evidence for these five features and explained how they work together to determine the fidelity and efficiency ofinitiation. A scanning mechanism for initiation can explain many of the effects of cap, context, position, etc . The scanning model (193) in its simplest form postulates that a 40S ribosomal subunit, carrying Met-tRNA;^" and an imperfectly defined set of initiation factors (302), enters at the 5' end of the mRNA and migrates linearly until it reaches the first AUG codon, whereupon a 60S subunit joins and the first peptide bond is formed . Evidence in support of the model has been adduced previously (62,193,197). More recent evidence for scanning includes the apparent queuing of 40S ribosomal subunits on long leader sequences (199) and the stalling of 40S subunits on the 5' side of a stable hairpin structure introduced between the cap and the AUG codon (195). The possibility of initiation by a mechanism other than scanning has been proposed (158) and is evaluated elsewhere (197) .The trick to identifying elements in 5' noncoding sequences that can modulate translation was to isolate each feature (200), an approach made possible by the techniques ofgenetic engineering . For example, by devising a transcript in which the first AUG codon was in an unfavorable context and hence "leaky," we were able to show that downstream