Recent studies of eyewitness memory have observed deleterious effects of producing a verbal description on later identification accuracy of a previously viewed face, an effect termed “verbal overshadowing” (Schooler & Engstler-Schooler, 1990). The current research investigated whether the phenomenon of verbal overshadowing may be constrained by variation in participants’ initial retrieval processes, such that verbalization of a previously viewed stimulus could produce either positive or negative influences on subsequent attempts at recollection. To assess the validity of this hypothesis, we manipulated participants’ response criterion during the verbal description task. As predicted, variation in response criterion significantly influenced not only the quality of the description generated, but also accuracy on a subsequent identification task. This retrieval-based effect was found to persist despite either a post-description delay (Experiment 1), or a source monitoring instruction at the time of the identification task (Experiment 2). We conclude that retrieval-based processes exert a powerful influence over the accuracy of verbalization and subsequent identification of a target face.