After viewing a crime, witnesses are frequently asked by police investigators to give a verbal description of the perpetrator. At a later time, witnesses may be asked to try to recognize the perpetrator in a lineup or a mugfile. The purpose of this research was to determine whether performance on a later recognition test is influenced by an earlier verbal description test of the target. Participants viewed six target faces, and after each, performed one of four post-exposure activities. Two were verbal description tasks. The descriptor checklist task had participants indicate, using a list of potential adjectives, those descriptors that best described the most recently seen face. The descriptor generate task had participants recall descriptors that best described the most recently seen face. Both description tasks were accomplished using a questionnaire that was identical except for the presence or absence of specific adjectives. Other participants were instructed to image the target face or they performed an irrelevant task during the post-exposure period. In a later recognition test, participants tried to find the targets in a set of faces containing 134 distractors. The results showed that participants using the checklist produced lower performance on the recognition test than participants who generated their own descriptors. Imaging produced the highest recognition performance but was not significantly different from the irrelevant condition. The recognition decrement of the descriptor checklist is explained in terms of interference caused by the presence of irrelevant face descriptors that added noise to participants' memory of the targets. Implications for testing witness memory by police investigators are discussed. For example, if verbal descriptions of faces are requested from eyewitnesses, descriptor recall may be preferred over a descriptor recognition because the former task does not degrade later recognition of the perpetrator.