Facial Composites and Eyewitness Memory: Accuracy and Limitations
Author: Ukrainian Psychological Hub · Published: October 11, 2026 · Editorial Policy
Facial composites are images assembled from a witness's memory of a person seen during an event. In criminal investigations, an artist or a computer-assisted system helps the witness construct a likeness that investigators may circulate to find leads. The central psychological problem is demanding: a witness must recover information about an unfamiliar face, translate that information into an image, and decide whether the developing picture resembles a memory that may already be incomplete. The final image may then influence other people's decisions.
Research establishes a useful distinction. A composite can be successful as an investigative lead when someone familiar with the depicted person recognizes it, yet its resemblance to the actual person can be imperfect. Conversely, a witness may feel satisfied with a composite even though independent observers struggle to identify whom it depicts. Neither outcome by itself establishes an individual's guilt. A research synthesis of facial-composite construction explains the purposes, varieties, and limitations of the method.
This article examines how facial composites work, what the experimental evidence actually measures, whether making or viewing one changes eyewitness memory, and how police, experts, attorneys, and courts should interpret the resulting evidence.
What Is a Facial Composite in Forensic Psychology?
A facial composite is a reconstructed likeness of a person's face, commonly created from a witness's recollection with help from a sketch artist, an image-part selection program, or a whole-face computer system. It is a product of a memory-retrieval and image-construction procedure. Its evidentiary status depends on how it was produced, what it is offered to show, and the procedural law of the relevant jurisdiction.
The usual investigative purpose is to elicit recognition by somebody who knows the person depicted. A circulated image might prompt a relative, coworker, or acquaintance to provide a name or other lead. Investigators must then assess that lead using independent evidence. The composite itself is a reconstruction, whereas a surveillance photograph records image information from a camera. Computer-generated or artist-generated detail can exceed what the witness actually remembered.
A composite is also distinct from a forensic facial comparison conducted between existing images, an automated face-recognition search, a witness's original verbal description, and a controlled lineup identification. Those methods have different inputs, error mechanisms, and validation requirements. Confusing them gives a constructed face a level of apparent objectivity that the procedure cannot supply.
The U.S. National Institute of Justice's eyewitness evidence guide treats developing and using composite images as a separate investigative procedure, alongside witness interviews and later identification procedures. It is a procedural reference published in 1999, so agencies should also consult current local policy and later research.
How Police Facial Composites Are Made
Forensic sketches and artist-led interviews
In an artist-led procedure, the witness describes facial characteristics while an artist draws and revises an image. The process may incorporate face shape, hair, eyebrows, eye placement, nose, mouth, facial hair, scars, or distinctive marks. Interview quality matters because questions can channel attention toward particular details, and some details may be reconstructed through inference rather than recalled from the original view.
An effective interviewer starts with an open narrative, separates remembered details from uncertainty, avoids suggesting features, and documents significant changes. The artist's skill and the witness's opportunity to observe both matter, but they represent different sources of information. A polished image can be persuasive without being accurate.
Feature-based systems
Traditional feature-based programs let witnesses select and adjust individual eyes, noses, mouths, hairstyles, and other facial components. Examples discussed in the literature include E-FIT and PRO-fit. This offers systematic control over details, but the procedure often demands verbal recall of features from a face originally processed as a whole.
The difficulty is especially pronounced for unfamiliar faces. A person can recognize an acquaintance effortlessly while struggling to describe the shape of that person's nose. A witness who cannot articulate a feature accurately may select an available component that looks approximately right, thereby introducing a detail that was never firmly represented in memory.
Holistic or whole-face systems
Holistic composite systems instead emphasize recognition and repeated selection among complete faces or face regions. Systems such as EvoFIT and EFIT-V use whole-face judgments and iterative modifications to approach the witness's recollection. These methods were developed partly because face recognition depends on configurations and relationships among features as well as individual parts.
The distinction is relative rather than absolute: feature-based systems display a whole face during editing, and holistic systems may permit changes to individual regions. Their procedures differ in which judgments they ask of the witness and in the order those judgments are made. The 2023 academic review of facial composites places these technologies in their historical and experimental context.
Digital and AI-assisted faces
A modern interface may generate plausible facial imagery quickly, yet photorealism is a property of the rendering system, not proof of memory accuracy. AI-generated facial detail may be especially compelling because the software fills visual gaps coherently. Unless a specific system has been independently validated on appropriate eyewitness populations and real-world conditions, claims about improved identification accuracy remain preliminary.
Investigators should preserve the original witness description, intermediate versions, operator choices, prompt and edit history where applicable, and the final image. Image generation should not silently convert guesses into purported observations. A digital composite's similarity to an actual photograph also cannot be treated as a calibrated probability of identity without separate validation.
Why Reconstructing an Unfamiliar Face Is So Difficult
Encoding happens before the witness is interviewed
Memory begins with the original viewing conditions. Exposure duration, viewing angle, lighting, distance, attention, obstructions, changing expressions, and distraction influence what information is encoded. Memory reconstruction cannot recover details that were never adequately perceived. The National Academies' review of basic eyewitness vision and memory research documents why identification depends on both visual conditions and later memory processes.
The witness may also have concentrated on an event, an action, or an immediate safety concern rather than the face. This affects available information without determining whether the witness was truthful or attentive in any global sense. A detailed narrative of the event and a weak face memory can coexist.
Face recognition and feature description use different demands
People commonly process faces configurally, integrating feature positions, proportions, and the appearance of the whole. Verbal descriptions force more attention onto discrete components. This creates a retrieval mismatch: recognizing a face from alternatives can be easier than producing an accurate description of each component independently.
Investigators therefore should distinguish the amount of detail a witness can verbally recall from the reliability of an eventual composite. More features do not necessarily produce a more recognizable face if their arrangement or selection is wrong.
Retrieval is reconstructive
A composite construction session is a sequence of memory judgments interleaved with external visual examples. When a witness selects a nose, changes a hairstyle, or approves an edit, that act may reflect genuine recollection, an approximate similarity judgment, a preference among limited choices, or a response to feedback. The process should permit uncertainty instead of forcing a definitive selection.
A clear warning that some displayed features are only options can help frame the task properly. However, research on memory contamination also means that an interviewer should avoid treating later certainty as proof that a feature came from the witnessed event.
Delay matters, but it has limits as a rule
Longer delays typically reduce the availability of some information about unfamiliar faces. In the 2015 regression and meta-analysis of facial composite naming, composites made after one or two days tended to be less identifiable than those constructed within approximately the first few hours. A 2025 experiment likewise found better naming for certain feature-based composites made after three to four hours than after two days.
These findings support timely, appropriately conducted interviewing. They do not create a fixed expiration date for memory, establish that every immediate session is best, or override a witness's health, fatigue, distress, and willingness to participate. The duration and timing of a session must be documented rather than inferred from an apparent image quality.
How Accurate Are Facial Composites?
Accuracy has several meanings in this field, and research findings change substantially depending on the measure selected.
Correct naming by people familiar with the target
A common experimental design asks one group of participants, who do not know the target, to construct composites after seeing a face. Another group, familiar with the target identity, attempts to name those images. The correct-naming rate measures whether the composite supplies recognizable cues to people who already know the person.
The Frowd and colleagues' 2015 synthesis aggregated 23 studies, including 6,464 naming responses across 1,069 participants. Under the conditions studied, composites constructed with the holistic EvoFIT system were more than four times as identifiable as those from the compared feature-based and sketch methods. That is a comparison within a particular research corpus, not a universal error rate for every police system or witness.
One 2013 experimental study reported a mean correct-naming rate of 74% when several carefully combined whole-face procedures were used. The figure came from its specific experimental procedures, target selection, and recognizers. It should never be described as the probability that a composite produced in an actual crime investigation is accurate.
Likeness ratings
Researchers may ask evaluators how closely the composite resembles a photograph of the target. Such ratings can reveal differences between methods or witness groups, but they are subjective assessments. A higher resemblance score does not directly convert into a chance that the target will be named or an innocent person falsely identified.
A study comparing adults and children found that adult constructors generally produced higher-rated holistic composites, and that children sometimes overestimated their own composites' likeness. The same paper examined later lineup performance and found a more concerning pattern for children, underscoring why appearance quality and witness memory must be analyzed separately. See Davis and colleagues' adult-child comparison.
Investigative leads and field outcomes
In real investigations, success may be reported as tips generated, possible suspects identified, arrests, or cases progressed. These outcomes depend on publicity, distribution channels, witness characteristics, the actual offender's social visibility, other investigative evidence, and the standard used to count a success.
A person may recognize a composite without being correct. An investigation may solve a case with help from a composite without proving that the likeness itself was accurate. Field success rates are consequently hard to compare across agencies and should be interpreted with denominators, independent verification, and clear outcome definitions.
False identifications, false alarms, and calibration
A facial composite can resemble multiple people. If a responder names an innocent person, that is an investigative false lead. A composite's naming rate and false-lead rate need to be considered together. Studies that only report correct target names cannot establish an overall risk of accusing innocent people.
There is no generally validated single probability score attached to an ordinary hand-drawn or computer-constructed facial composite. Software resemblance scores are also system-specific. Calibration would require demonstrating how stated probabilities or score bands correspond to true identity outcomes in representative conditions. A plausible face, high witness satisfaction rating, or numerical similarity score is not evidence of such calibration.
Do Holistic Composite Systems Outperform Sketches and Feature-Based Systems?
A substantial research literature favors whole-face construction for the limited outcome of correct composite recognition or naming. The 2015 meta-analytic comparison and the 2023 review chapter both explain why aligning the construction task with face-recognition processes can improve recognizability.
Research also indicates that the interview and construction process should be matched cognitively. In an experimental series, whole-face or character judgments combined with corresponding feature-selection strategies produced better-identified composites than mismatched procedures. See the study on cognitive alignment.
However, a method that makes a likeness easier to name has not thereby demonstrated a lower risk of error in every subsequent lineup, a better ability to distinguish the culprit from innocent candidates, or superior performance in every age and cultural group. Systems should be evaluated by outcome, setting, operator training, witness population, and the degree to which studies approximate actual investigations.
Does Creating a Facial Composite Change Eyewitness Memory?
This question concerns a different outcome from the quality of the finished image. The witness who produces a composite may later be asked to identify somebody in a lineup. During image construction, the witness sees many artificial faces or facial features, makes repeated choices, and may become familiar with the product of those choices. Researchers have therefore investigated whether the activity helps, harms, or leaves subsequent identification unaffected.
The strongest synthesis finds no consistent overall impairment
A 2021 meta-analysis by Tredoux and colleagues combined 23 studies, 56 effect sizes, and 2,276 participants. It found no statistically significant overall negative effect of constructing a composite on later lineup identification. It reported some weak evidence of reduced incorrect identifications in target-present lineups and found no effects in its signal-detection measures. The authors identified continuing needs for research that better matches real investigations.
This synthesis supports a carefully bounded conclusion: existing aggregated experiments do not establish that constructing facial composites generally worsens eyewitness identification accuracy. It does not prove that every construction method is harmless, that every procedural variation is equivalent, or that effects cannot differ for specific groups and delays.
Individual studies produce different results
In a 2019 experiment using more realistic delays, participants viewed a staged crime, constructed a feature-based or holistic composite, and later completed an identification task. Neither construction method significantly improved or impaired subsequent identification relative to the comparison condition. The authors distinguished their results from shorter-delay research that had reported benefits.
A study of adults and children found poorer correct identification among child composite constructors than child controls, whereas the comparable adult difference was absent. This is an important population-specific finding. It calls for careful child-witness protocols and further replication rather than automatic application of adult findings to children.
Constructing a composite and viewing somebody else's composite are different exposures
The witness who helped build a composite knows it is an approximate product of their own choices. A different witness who views the picture may receive a new and potentially misleading source of facial information without that procedural context. Treating both exposures as the same activity would conceal a meaningful difference.
A comprehensive 2020 review of exposure to facial composites found mixed results: misleading composites sometimes damaged later identification, especially when an innocent suspect shared the composite's features; better composites produced positive or null effects in some studies. The review could not perform a full formal meta-analysis because of limitations in reported statistical information.
The practical implication is to protect witnesses from exposure to one another's composites before independent memory accounts and identifications are obtained. This is a precaution grounded in evidence of possible contamination, not proof that every exposure will alter memory.
Can a Facial Composite Contaminate a Witness's Memory?
Memory contamination arises when information encountered after the event becomes incorporated into later recollection, affects familiarity, or influences a decision. A composite may create new opportunities for confusion because its features look like remembered perceptual information. A witness who later recognizes a face may be responding partly to resemblance to the composite rather than to the originally observed person.
Source monitoring is central. Investigators should ask whether a particular detail originated in the event, an interview, an image displayed during construction, a social-media post, or another person's account. Post-event familiarity can be subjectively vivid while having a different evidentiary origin from direct observation.
Contamination risk grows when an image is repeatedly displayed, an operator strongly endorses one version, witnesses collaborate, or a composite is circulated widely before witnesses complete independent identification procedures. These are reasons to document the information environment and separate witnesses, not grounds to pronounce a particular identification false without case-specific analysis.
A key finding in the Sporer and colleagues' review is that a misleading composite may be especially problematic if it shares features with an innocent suspect. This mechanism matters even when a composite has genuine investigative utility.
Can the Cognitive Interview Improve a Composite?
A cognitive interview generally uses memory-supporting techniques such as open recall and appropriate contextual reinstatement while avoiding leading questions. For facial composites, researchers have adapted it with a holistic interview phase, sometimes asking witnesses to think about overall facial impressions after their initial description.
In a 2008 study of the holistic cognitive interview, the resulting feature-based composites were named more often than composites produced after a standard cognitive interview. The specific manipulation involved global impressions and multiple holistic judgments, and the outcome was composite recognition rather than criminal case resolution.
A 2013 whole-face study found gains from combining interviewing, selection, and presentation manipulations. The 2025 feature-composite study likewise reported benefits in certain short-delay conditions when holistic trait-recall techniques and an adjusted viewing method were combined.
The mechanism is best understood as helping witnesses retrieve and organize visual information in a manner more compatible with recognition. Researchers' use of personality-like impressions as prompts does not imply that a witness can infer intelligence, aggression, honesty, or criminal propensity accurately from facial appearance. Such judgments are memory-retrieval prompts in a controlled method, not psychological assessments of the person depicted.
What Factors Influence Facial-Composite Accuracy?
Observation quality and familiarity
A face observed briefly, at a distance, or under poor lighting may yield a weak representation. Repeated prior familiarity normally supplies more robust information than a single encounter. Age, attention, viewing angle, and occlusion also matter, and their effects should be analyzed in the specific eyewitness task rather than combined into a single intuitive reliability score.
Time, intervening information, and repeated retrieval
Elapsed time matters alongside the events that occur during that interval. Discussions with others, viewing photos, searching online, or exposure to publicity can change the information available to the witness. A repeated construction attempt may introduce new external details as well as offer opportunities for recollection. Investigators need a record of both timing and intervening exposures.
Construction method and operator influence
Programs differ in their stimulus sets, editing capabilities, construction order, and the degree to which an operator can shape the result. Interviews differ in training and safeguards. The same witness might create different images with different interfaces; this variance is an important limit of the method rather than evidence that the witness is dishonest.
Age, development, and accessibility
The child-witness literature cautions against assuming adult effects generalize across developmental groups. Accessibility adjustments for visual, cognitive, language, or communication needs should preserve the witness's independent recollection and be documented. A witness's ability to navigate software is not a measure of the quality of their original memory.
Race, ethnicity, and population representation
Research on eyewitness recognition has documented population-dependent performance and cross-race identification effects, but findings on lineup recognition cannot be mechanically converted into facial-composite error rates. A composite system's face database may underrepresent particular populations or provide unsuitable feature choices. Inclusive stimulus libraries, independent validation, interpreter quality, and culturally competent interviewing are relevant quality controls. Broad racial labels alone are inadequate measures of an individual witness's accuracy.
Publication format and audience
Image resolution, hair and clothing presentation, cropping, facial expression, and the context in which a composite is shown may affect recognition. A familiar observer and an unfamiliar observer perform different tasks; an unfamiliar viewer cannot normally supply a valid identity simply by judging that the image resembles someone seen online.
How Should Investigators Use Facial Composites?
A defensible investigative workflow begins with an independent initial account, records observation conditions and uncertainty, and uses an appropriately trained composite operator. The session should document prompts, witness-initiated changes, system selections, dates, and the image versions retained. The witness should understand that the goal is an approximate lead-generating likeness.
The National Institute of Justice's law-enforcement guide includes specific sections on preparing and using composites, instructing witnesses, and recording the procedure. Its companion trainer's manual supports training and documentation. Both are historical official U.S. resources and must be considered with current agency requirements.
When a composite is publicized, authorities should describe it as an approximate representation made from a witness's memory, identify its intended investigative use, and use an official channel for tips. Investigators must independently corroborate leads and minimize reputational harm to people who resemble the picture.
Other eyewitnesses should give independent descriptions and make identification decisions without first encountering the composite when that can be avoided. If exposure has occurred, its timing and content should be recorded. The original witness's future identification should follow appropriate lineup safeguards instead of an informal comparison with the composite.
How Should Courts and Forensic Experts Interpret Composites?
A constructed portrait is the output of a process, not a photograph of a person at the crime scene. A forensic expert may explain relevant research, how a particular method was administered, known uncertainty, and the limits of the studies. The fact finder applies the jurisdiction's admissibility and evidentiary standards. Psychological findings about face memory cannot themselves decide whether an image is admissible or whether a defendant committed a crime.
The National Academies' assessment of eyewitness identification emphasizes the need to evaluate how eyewitness evidence is collected and interpreted. Many courtroom disputes concern lineups, confidence, and identification procedures; composite construction raises related but distinct questions. A composite may be offered to show investigative steps, the witness's earlier description, or another legally relevant issue. The reason it is offered affects which legal rules apply.
If a witness later identifies a suspect, the expert should distinguish the original viewing opportunity, the composite construction, subsequent exposure, and the identification procedure. The expert should also ask whether the suspect was already known to investigators and whether any potentially suggestive contact occurred. A good match between the image and a charged defendant can be assessed retrospectively, but that match does not establish that the defendant was the originally seen person.
Terms such as forensic opinion, legal finding, identification decision, and guilt belong to different domains. A specialist's account of research is not a judicial finding. A witness's likeness judgment is not a calibrated identification probability. A laboratory effect size does not supply the probability of error in one trial without the relevant case-specific information and a defensible statistical model.
Limitations of the Evidence Base
Controlled studies permit researchers to isolate construction systems and interview variations. They also often use photographs, short videos, known targets, familiar celebrity names, or student volunteers. These materials differ from real witnessing conditions, especially with respect to stress, time, movement, unfamiliarity, and the consequences of making a mistake.
Correct composite naming is easier to study than the full downstream chain from construction to tip, corroboration, lineup, charge, and verdict. Many studies therefore evaluate an intermediate outcome that is useful but narrower than case-solving accuracy. Likewise, a large percentage improvement over a weak comparison condition may coexist with a modest absolute recognition rate.
Several published findings depend on particular composite platforms and investigator training. They do not amount to proof that all systems marketed as holistic, all generative AI interfaces, or all sketch artists perform alike. Technology-specific validation, preregistered replication, field auditing, and reports of incorrect leads would strengthen the evidence.
Meta-analyses provide more dependable averages but cannot remove every limitation of their component studies. The 2021 synthesis explicitly calls for more ecologically valid work, while the 2020 composite-exposure review points to reporting limitations that prevented more comprehensive pooled analysis. Conclusions should track the exact comparison being studied.
Common Myths About Police Sketches and Facial Composites
A detailed composite is necessarily an accurate one
Photorealistic finishing, precise hair texture, and well-balanced features can convey confidence. The evidence instead concerns the fidelity of underlying memory and the construction procedure. A plausible face can contain invented or contaminated details.
A recognizable composite proves the person recognized is guilty
Recognition provides a lead. Establishing that the recognized individual was present or responsible requires independent investigation and application of the relevant legal standards. An innocent person may resemble the image.
Creating a composite always damages eyewitness memory
The 2021 meta-analysis did not find a consistent overall detrimental effect of construction. Particular populations and conditions may still show differences, and viewing someone else's misleading composite is a separate exposure.
A newer algorithm automatically solves witness-memory errors
An improved generator may make images more visually coherent or easier to edit. The witness's perceptual and memory limits remain, and the generated face may add detail unsupported by recollection. Claims about predictive performance require direct validation.
The witness's confidence in the picture establishes its accuracy
Confidence that the final image looks right is a judgment about the construction process. Recognition by independent observers, later lineup identification, and legal proof are different outcomes. None should be substituted for another.
Frequently Asked Questions
What is a facial composite used for?
It helps generate investigative leads by depicting a remembered face in a form that people familiar with the individual may recognize. It is an approximate reconstruction used during an inquiry, and any resulting identification must be checked independently.
How accurate are police composite sketches?
There is no universal accuracy percentage. Results vary with the witnessing conditions, delay, interview, system, target familiarity, and measure of success. Laboratory correct-naming results cannot be applied as the error probability of an individual police sketch.
Are computerized composites better than hand-drawn sketches?
Some evaluated holistic computer systems have produced more recognizable composites than certain feature-based and sketch methods in the studies reviewed in the 2015 meta-analysis. That comparison is method-specific and does not rank every individual artist or product.
Does making a composite improve the witness's memory?
The evidence does not demonstrate a consistent improvement in later lineup identification. A 2021 meta-analysis found no reliable overall beneficial or detrimental effect. Particular study designs have found gains, null effects, or population-specific costs.
Can viewing a composite cause a false identification?
It can in some circumstances. A research review reports negative effects particularly when a misleading composite resembles an innocent suspect. Other studies find no effect or benefits from accurate composites. Independent witness procedures and exposure records remain important.
Why can someone recognize a face but fail to describe it?
Recognizing a whole face and verbally recalling each facial feature place different demands on memory. Composite systems that emphasize whole-face selection seek to align the reconstruction task with ordinary facial recognition.
Can facial composites be used in court?
Their admission and permitted uses depend on the jurisdiction, the purpose for which they are offered, and the case's procedural circumstances. Psychological research informs evidentiary evaluation; courts determine admissibility and legal weight.
Can AI produce a reliable facial composite from a witness's description?
AI can produce a visually plausible image. Reliability for identifying the actual person requires independent empirical testing on suitable tasks, witnesses, populations, and operational conditions. The algorithm's appearance quality does not substitute for such evidence.
Should witnesses see a published composite before a lineup?
Investigators should ordinarily protect the independence of eyewitness accounts and identification decisions. If a witness has seen the composite, that exposure should be documented and considered when interpreting the later decision. Specific procedures are controlled by current local law and policy.
For broader context, see Eyewitness Testimony: How Memory Works in Legal Cases.
Related Articles
References
Davis, J. P., Thorniley, S., Gibson, S., & Solomon, C. (2016). Holistic facial composite construction and subsequent lineup identification accuracy: Comparing adults and children. Journal of Psychology, 150(1), 102–118. https://doi.org/10.1080/00223980.2015.1009867
Frowd, C. D., et al. (2008). Improving the quality of facial composites using a holistic cognitive interview. Journal of Experimental Psychology: Applied. https://pubmed.ncbi.nlm.nih.gov/18808281/
Frowd, C. D., et al. (2013). Whole-face procedures for recovering facial images from memory. Science & Justice, 53(2), 89–97. https://doi.org/10.1016/j.scijus.2012.12.004
Frowd, C. D., et al. (2015). A decade of evolving composites: Regression- and meta-analysis. Journal of Forensic Practice, 17(4), 319–334. https://doi.org/10.1108/JFP-08-2014-0025
Frowd, C. D., Portch, E., Estudillo, A. J., Ford, C. J., Purcell, A., Pitchford, M., & Brown, C. (2025). The value of whole-face procedures for the construction and naming of identifiable likenesses for recall-based methods of facial-composite construction. Applied Cognitive Psychology, 39(4), e70015. https://doi.org/10.1002/acp.70015
National Institute of Justice. (1999). Eyewitness evidence: A guide for law enforcement. U.S. Department of Justice. https://nij.ojp.gov/library/publications/eyewitness-evidence-guide-law-enforcement
National Institute of Justice. (2003). Eyewitness evidence: A trainer's manual for law enforcement. U.S. Department of Justice. https://nij.ojp.gov/library/publications/eyewitness-evidence-trainers-manual-law-enforcement
National Research Council. (2014). Identifying the culprit: Assessing eyewitness identification. National Academies Press. https://doi.org/10.17226/18891
Pike, G. E., Brace, N. A., Turner, J., Ness, H., & Vredeveldt, A. (2019). Advances in facial composite technology, utilizing holistic construction, do not lead to an increase in eyewitness misidentifications compared to older feature-based systems. Frontiers in Psychology, 10, 1962. https://doi.org/10.3389/fpsyg.2019.01962
Sporer, S. L., Tredoux, C. G., Vredeveldt, A., Kempen, K., & Nortje, A. (2020). Does exposure to facial composites damage eyewitness memory? A comprehensive review. Applied Cognitive Psychology, 34(5), 1166–1179. https://doi.org/10.1002/acp.3705
Tredoux, C. G., Frowd, C., Vredeveldt, A., & Scott, K. (2023). Construction of facial composites from eyewitness memory. In Biomedical Visualisation, Vol. 13, pp. 149–190. Springer. https://doi.org/10.1007/978-3-031-13021-2_8
Tredoux, C. G., Sporer, S. L., Vredeveldt, A., Kempen, K., & Nortje, A. (2021). Does constructing a facial composite affect eyewitness memory? A research synthesis and meta-analysis. Journal of Experimental Criminology, 17(4), 713–741. https://doi.org/10.1007/s11292-020-09432-z