Errors during the process of translating mRNA information into protein products occur infrequently. Frameshift errors occur less frequently than other types of errors, suggesting that the translational machinery has more robust mechanisms for precluding that kind of error. Despite these mechanisms, mRNA sequences have evolved that increase the frequency up to 10,000-fold. These sequences, termed programmed frameshift sites, usually consist of a heptameric nucleotide sequence, at which the change in frames occurs along with additional sequences that stimulate the efficiency of frameshifting. One such stimulatory site derived from the Ty3 retrotransposon of the yeast Saccharomyces cerevisiae (the Ty3 stimulator) comprises a 14 nucleotide sequence with partial complementarity to a Helix 18 of the 18S rRNA, a component of the ribosome's accuracy center. A model for the function of the Ty3 stimulator predicts that it base pairs with Helix 18, reducing the efficiency with which the ribosome rejects erroneous out of frame decoding. We have tested this model by making a saturating set of single-base mutations of the Ty3 stimulator. The phenotypes of these mutations are inconsistent with the Helix 18 base-pairing model. We discuss the phenotypes of these mutations in light of structural data on the path of the mRNA on the ribosome, suggesting that the true target of the Ty3 stimulator may be rRNA and ribosomal protein elements of the ribosomal entry tunnel, as well as unknown constituents of the solvent face of the 40S subunit.