More than two decades ago Marilyn Kozak proposed the scanning model of tran
slation initiation, whereby translation is initiated at the first AUG codon
that is in a particular context. In this article, we re-examine the contex
t of initiator codons using a large dataset of curated human transcripts. W
e find that more than 40% of transcripts contain AUG codons upstream of the
actual start codon and that most authentic AUGs contain three or more mism
atches from the consensus sequence, CCACCaugG. Also, in a large fraction of
transcripts, the sequences surrounding the initiator codon deviate more fr
om the consensus than those surrounding upstream AUGs, indicating that tran
slation initiation from downstream AUGs is more common than generally belie
ved.