The expression levels of coding sequences for pituitary growth hormone, introduced into Escherichia coli by genetic manipulation techniques, vary markedly according to the precise sequence introduced. In order to understand the basis of this variation more fully, we have studied the relationship between the level of expression in E. coli of a series of ovine growth hormone variants and the nucleotide sequences coding for their N-terminal regions. Sequence variation resulted from the introduction of deletions, or site-directed mutations, into a plasmid containing the coding sequence for ovine growth hormone preceded by the initiation codon and 25 bases derived from β-galactosidase or linker regions of plasmid pUC8. The expression levels of the variants varied from less than 0.01% to over 34% of the total cell protein, indicating that changes in the nucleotide sequence close to the initiation codon had a marked effect on expression level. The results of a comparison of closely related sequences in pairs of plasmids giving poor or good expression are consistent with the hypothesis that poor translation of growth hormone mRNAs is caused by the presence of secondary structures close to the initiation codon. Secondary structures are identified that appear to explain the variation in expression levels.