When eyewitnesses identify someone in a police lineup, it is common practice to ask them for their confidence in that decision. In light of research pointing to reactivity effects of confidence judgments in various cognitive tasks, the question arises whether obtaining confidence judgments during police lineups affects the cognitive processes underlying eyewitness responses. These processes were measured using the two-high threshold eyewitness identification model. Participants watched a staged-crime video before completing four sequential lineups, one for each culprit. Asking participants for their confidence after each decision in the lineup had no effect on any detection-based processes. Instead, it significantly increased the probability of guessing-based selection. This pattern can be explained by assuming that allowing eyewitnesses to report low confidence for identifications makes it feel acceptable to identify someone even when the decision is based on guessing. Supporting this interpretation, instructing participants to identify lineup members only with high confidence and denying them the opportunity to report low confidence decreased guessing-based selection relative to a control condition without confidence judgments. These findings indicate that the effect of confidence judgments on eyewitness identification depends on how confidence is used: The opportunity to qualify an identification with a low-confidence judgement enhances guessing-based selection, whereas restricting identifications to high-confidence decisions reduces it relative to a condition without any confidence judgements. For police practice, these results suggest that confidence judgements during lineups should be used in combination with an explicit instruction for eyewitnesses to make an identification only when they are very confident.