The identification procedure can greatly affect eyewitness performance, but this may be contingent upon a relatively weak memory for the perpetrator. In a large preregistered experiment (N = 13,728), we manipulated memory strength and tested participants with a target-present or -absent showup or lineup (size 3 or 6). All fillers were description-matched but were of low or high similarity with the target. We replicated the advantage of fair simultaneous lineups over showups and the advantage of low- over high-similarity fillers when memory for the perpetrator's face was weaker (short exposure duration), but both effects were significantly reduced when memory was stronger. There was no effect of lineup size regardless of memory strength or filler similarity. We conclude that some recommendations to police may be more robust than others across changes in estimator variables such as memory strength and that more research is needed on interactions between estimator and system variables.