Oligonucleotides designed to create hairpin structures were inserted upstream from the ATG initiator codon in several plasmids that encode preproinsulin, and the effects on translation were monitored in COS cells transfected by the vectors. Creation of a hairpin (delta G = -30 kcal/mol) that directly involves the ATG triplet at the start of the preproinsulin coding sequence does not reduce the yield of proinsulin. However, a more stable stem-and-loop structure (delta G = -50 kcal/mol) reduces the proinsulin yield by 85-95%. The stable hairpin inhibits even when it occurs at the midpoint of the 5' untranslated sequence and thus involves neither the cap nor the ATG codon. Presumably the migrating 40S ribosomal subunit can melt moderately stable duplexes but stalls at structures (delta G = -50 kcal/mol) that resist unfolding. Other experiments argue against the idea that sequestering the 5'-proximal ATG codon in a hairpin structure might allow it to be skipped by ribosomes in favor of an exposed ATG triplet farther downstream: when the primary sequence around the first ATG triplet is favorable for initiation, no translation from a downstream site can be detected, irrespective of whether the first ATG codon is single-stranded or base-paired.