Purpose This study examined how constructing and viewing facial composites affects memory accuracy. Method Participants (n = 240) viewed a target face for 60 s; half were then randomly assigned to construct a facial composite while the other half completed a distracter task. Participants were then randomly assigned to view a composite either once, twice or not at all during a 1‐week delay between the first and second session. Those who constructed a composite viewed their own composite during the 1‐week delay, while those who did not construct a composite were yoked with someone who did, thereby viewing their counterpart's composite. In session 2, all participants completed a six‐person forced‐choice identification task. The participant was asked to identify the target from a six‐person photo array. The photo array was composed of the target, the composite, and four morphed images that combined certain percentages of both the target and composite. Results Results showed that identification accuracy was lower for those who created a composite versus those who did not. These individuals were likely to identify an image consisting of a blend between the target and composite more often than those not creating a composite. In addition, those who only viewed a composite experienced memory contamination similar to those who created a composite. Conclusion Our findings illustrate that creating a composite or simply viewing someone else's composite hinders identification accuracy and one's memory for the target face. This research provides important implications regarding the methods used in collecting eyewitness descriptions during the investigative process.