Using an interference paradigm, we demonstrate that there is a place for a direct coding mechanism in a comprehensive theory of frequency coding. Ss were presented words whose frequency was judged later. Under one set of instructions, these words were coded in terms of numerical associates; under another set of instructions, the coding was governed by nonnumerical associates. The condition using numerical associates resulted in frequency estimations that were of lesser quality than those produced in the control condition. This effect, moreover, was a function of the encoding of the target words, not just their retrieval.We thank David Bryant, Robert Greene, Robert Rose, and William Whitlow for their incisive comments on an earlier version of this article.