Skip to content
Search resources Open menu

Online Offense

Estimating Age from Images

Published

Snap Shot

  • The question: How reliably can anyone tell from an image whether the person depicted is under 18?
  • What the research shows: Images of clearly prepubescent children, and early pubertal development (Tanner stages 1 and 2), can support an opinion that a person is a minor. For post-pubertal subjects, studies have found frequent misclassification of adults as minors; in one small study (ten images of adult women selected from websites featuring youthful-appearing models, 16 pediatric endocrinologists as raters), 69 percent of the 160 judgments classified the women as under 18; the study did not measure how often minors are judged to be adults.
  • What it does not show: Tanner staging was not designed to estimate age; its developer warned against that use, and a systematic review found it ineffective for stages 3 through 5. Misclassification studies do not show that a particular disputed image depicts an adult; the uncertainty runs in both directions. Automated tools are triage aids, not proof of age.
  • The law: Federal courts generally allow juries to assess age from the images (Riccardi); Katz observed experts may be needed for post-pubertal, youthful-appearing subjects. Opinions based on specialized training must satisfy Rule 702. Arizona’s § 13-3556 inference is permissive.
  • Issues for each side: Prosecutors should charge images where minority is clear and support age with independent evidence such as victim identification or chat context; defense counsel should challenge Tanner-based or appearance-based opinions for post-pubertal subjects and seek identification or age documentation, recognizing these challenges do not reach clearly prepubescent images.

Every prosecution involving child sexual abuse material (CSAM) requires proof that a person depicted was a minor. Statutes and courts use the term “child pornography,” and this paper uses that term when describing the law; elsewhere it uses CSAM. When the person depicted is a young child, age is rarely disputed. When the person depicted appears to be an adolescent or a young adult, and no one has identified the person or established a date of birth, the question of age can become the central factual issue in the case.

Courts have heard age opinions from investigators, physicians, nurses, and other professionals, sometimes based on the stages of sexual development described by the pediatrician James Tanner, and sometimes based on general impressions of body size, facial features, and context. Researchers have tested how accurately such judgments can be made from photographs. Their findings show a clear pattern: estimating age from images of clearly prepubescent children is a different and far more reliable task than estimating whether a physically mature or nearly mature person is under or over 18.

This paper reviews the developmental science of puberty and its variability, the origin and intended purpose of Tanner staging, the empirical studies of expert, lay, and automated age estimation from images, and the legal standards that govern who may offer age testimony and whether a jury may decide age from the images alone.

Puberty and Its Variability

Puberty is a sequence of physical changes, including the development of breasts and genitalia, the growth of pubic hair, a growth spurt in height, and changes in facial structure and body composition. The sequence is relatively predictable, but its timing and tempo vary widely among healthy individuals. The onset of puberty in healthy girls spans several years, and the interval from onset to physical maturity also varies by years, so two people of the same chronological age may be at very different stages, and two people at the same stage may differ in age by several years. Variation is associated with genetics, nutrition, body mass, ethnicity, and environment.

Puberty has also shifted earlier over recent decades. A systematic review and meta-analysis of studies from around the world found that the average age at the onset of breast development in girls declined by about three months per decade from 1977 to 2013 (Eckert-Lind et al., 2020). Secular trends of this kind mean that reference data from older studies may not describe current populations, and data from one population may not apply to another. In a review prepared for pediatricians asked to give opinions about images of Southeast Asian girls, Stathopulu et al. (2003) found limited population-specific data on early pubertal development and noted differences in the interval between early breast development and menarche across populations.

A consequence of this variability is that the end of physical maturation is not tied to the 18th birthday. Many adolescents complete most visible pubertal development well before 18, and some adults retain features that are often associated with adolescence. The task of distinguishing a 16-year-old from a 19-year-old from physical appearance therefore asks the observer to make a distinction that biology does not reliably mark.

Tanner Staging: Purpose and Limits

Tanner staging, also called sexual maturity rating, describes pubertal development in five stages for breast development in girls, genital development in boys, and pubic hair in both sexes. Stage 1 is prepubertal and stage 5 is adult. The system was developed to describe and track the progress of puberty in clinical and research settings, where the person’s age is already known, and to identify early and late maturers.

In a 1998 letter to Pediatrics, Arlan Rosenbloom and James Tanner warned against using the stages to estimate chronological age, particularly in forensic cases involving photographs (Rosenbloom & Tanner, 1998). They explained that no equations exist for estimating age from stage, that the unreliability of staging would introduce large errors even if such equations existed, and that staging from nonstandardized photographs rather than clinical examination adds further unreliability. They cautioned physicians against offering expert opinions on chronological age based on Tanner stage. A responding letter argued that staging can still be informative about whether a person appears to be a minor in some cases (Kutz, 1999), and that disagreement frames the current understanding: staging can indicate that a person is prepubescent or early in puberty, but it cannot assign a chronological age, and it cannot distinguish older adolescents from young adults.

A 2021 systematic review reached the same conclusion. Reviewing published studies and case reports on the use of Tanner staging to estimate age in alleged child pornography cases, the authors concluded that the method is ineffective for persons in stages 3 through 5, and that in stages 1 and 2 it may support an opinion that the person is under 18 but cannot establish a specific age (Oliván-Gonzalvo et al., 2021).

Studies of Accuracy

Several studies have tested age estimation using images of adults whose ages were documented, which allows accuracy to be measured directly.

Cattaneo et al. (2009) asked groups of physicians (forensic pathologists, pediatricians, and gynecologists) and lay participants in Italy and Germany to judge whether 11 adult women in images taken from legal pornographic websites were under or over 18, based on breast development, pubic hair, and overall appearance. All of the women were adults. Many raters in every group judged some of them to be minors, and accuracy varied considerably by professional group and by country; forensic pathologists performed best, and pediatricians and gynecologists performed worse than lay raters, judging the adult women to be under 18 more often (in Italy, 73 and 69 percent, compared with 50 percent for lay raters; Oliván-Gonzalvo et al., 2021). The authors concluded that visual age assessment of post-pubertal individuals from images carries major uncertainty, particularly around the 18-year threshold.

Rosenbloom et al. (2012) independently classified breast development in published images of 547 adult women as Tanner stage 4 or stage 5. All four examiners, who were pediatric endocrinologists, agreed on stage 4 for only 17 images (3.1 percent), but at least one examiner rated 26.5 percent of the images as stage 4. The authors concluded that stage 4 breast development persists in a substantial number of adults and that the distinction between stages 4 and 5 is too unreliable, especially in nonclinical images, to support testimony that a person is a minor. In a related study, 16 pediatric endocrinologists judged ten images of adult women selected from websites that featured youthful-appearing models; 69 percent of the 160 judgments classified the women as under 18 (Rosenbloom, 2013).

Studies using facial images of known minors have produced more encouraging results within limited age ranges. In a pilot study of frontal facial photographs of 50 girls and young women aged 10 to 19, both experienced raters and an automated program estimated age with a mean absolute error of about one and a half years, and the authors concluded that facial age estimation can serve as a preliminary screening tool for juvenile images (Ratnayake et al., 2014). A related study using facial proportions from standardized photographs of subjects aged 6, 10, 14, and 18 correctly classified about 60 percent of cases into their age groups (Cattaneo et al., 2012). An average error of one and a half years is useful for screening large volumes of images, but it is large relative to the question of whether a person is 17 or 19, and these studies used standardized photographs rather than the varied lighting, angles, makeup, and editing found in real material.

Research with law enforcement coders shows that classifying images by age and content is difficult in practice. In a study in which an experienced researcher and police staff in the United Kingdom independently coded large sets of seized images, coders showed both substantial agreement and meaningful disagreement, and focus groups identified the developmental stage of the person depicted and ambiguity in the context of the image as the main sources of difficulty (Kloess et al., 2019).

Automated Age Estimation

Digital forensic examiners increasingly use software to sort large collections, including tools that estimate the age of faces. Research on these tools for CSAM investigations has found that commercial facial age estimation services trained largely on adult faces performed less accurately on images of children and adolescents, and that models trained specifically on underage faces improved accuracy (Anda et al., 2018, 2020). Image quality, pose, and other factors affect performance. Researchers in this area describe the tools as aids for triage, helping investigators prioritize review and identify possible victims, rather than as a means of establishing the age of a particular person in court. Published appellate decisions treating automated age estimates as substantive proof of age are scarce, and any such use would require the same showing of reliability, error rates, and fit to the images at issue that Rule 702 demands of other methods.

The Prepubescent Distinction

The limitations described above concern the estimation of age for people who are pubescent or post-pubescent. They do not apply in the same way to images of young children who are clearly prepubescent, whose small body size, proportions, and absence of pubertal development are unmistakable. The systematic review discussed above found that stages 1 and 2 can support an opinion that a person is a minor. In federal cases, prepubescent victims appear in nearly all non-production cases, and a majority involve images of infants or toddlers (U.S. Sentencing Commission, 2021). Disputes about age therefore arise in a minority of cases and concern a specific category of images, those depicting people who appear to be in mid-to-late adolescence or young adulthood. Other evidence, such as identification of a known victim through the National Center for Missing & Exploited Children’s Child Victim Identification Program, file names and descriptions, the context of the conversation in which images were exchanged, or identification of an adult performer, often resolves the question without reliance on appearance.

Limits of the Evidence

The research base on age estimation from images is small. Most studies used a limited number of images, often of adult women selected for youthful appearance, and few included males or a range of ethnic backgrounds. Studies that used images of adults selected for youthful appearance measure the risk of misclassifying adults as minors; they do not measure how often minors are misclassified as adults, which is the error of concern to prosecutors and child-protection agencies. Studies with standardized photographs of known minors show what is possible under ideal conditions, not under the conditions of real material. The automated tools are developing quickly, and published accuracy figures may not describe current software.

No published method can reliably determine the age of a particular post-pubertal person from a particular image alone. Research can describe the reliability of a method and the range of error to be expected, and it can identify the conditions under which estimates are more or less trustworthy. Whether a person depicted was a minor remains a question of fact for the jury, to be decided on all the evidence.

Questions to Ask About This Research

When a witness cites the finding that 69 percent of 160 judgments classified adult women as under 18 (Rosenbloom, 2013), ask who was studied: the images were of adults selected from websites featuring youthful-appearing models, so the figure measures one direction of error and does not show how often minors are judged to be adults or that a disputed image depicts an adult. Ask also how much evidence it rests on, since it involved ten images and 16 raters. When a witness offers a Tanner-based opinion that a post-pubertal person is a minor, ask whether the conclusion follows from the method’s design, because Tanner staging was built to track puberty when age is already known and was found ineffective for stages 3 through 5 (Oliván-Gonzalvo et al., 2021). When a facial estimate with a mean absolute error of about one and a half years (Ratnayake et al., 2014) is offered, ask what that error means for whether this person was 17 or 19, and whether standardized photographs resemble the image at issue. These questions are explained in “Reading Research Claims: A Guide for Attorneys” on this site.

Age Determination and the Law

Federal and Military Standards

The government must prove beyond a reasonable doubt that a person depicted was a minor, and under United States v. X-Citement Video, Inc. (1994) it must also prove that the defendant knew the person depicted was a minor. Federal courts have generally held that expert testimony is not always required and that a jury may determine whether a person depicted is a minor by viewing the images, drawing on common experience. In United States v. Nolan (1987), the First Circuit held that the factfinder could infer from the images themselves that real children were depicted. The Tenth Circuit in United States v. Riccardi (2005) affirmed a conviction in which the district court admitted images of subjects the jury could assess from common experience and excluded others depicting subjects close to 18; in United States v. Kimler (2003), it held that Ashcroft v. Free Speech Coalition (2002) did not require expert testimony that images depicted real children. In United States v. Broyles (1994), the Eighth Circuit found the evidence sufficient where the jury had the video, other testimony, and an expert opinion. The Fifth Circuit in United States v. Katz (1999) took a case-by-case approach, observing that expert testimony may be unnecessary when the subjects are obviously young but may be necessary when the subjects are post-pubertal and appear youthful, and affirming the exclusion of images where the government’s expert, relying on Tanner staging, could not reliably establish age.

When a witness does offer an opinion on age, the rules of evidence determine its form. A lay witness may give an opinion under Federal Rule of Evidence 701 only if it is rationally based on the witness’s perception, helpful, and not based on scientific, technical, or other specialized knowledge within Rule 702. An investigator’s or physician’s opinion that rests on training in pubertal development, Tanner staging, or experience reviewing images is specialized and must satisfy Rule 702. As amended December 1, 2023, Rule 702 requires the proponent to show by a preponderance of the evidence that the opinion rests on sufficient facts or data, is the product of reliable principles and methods, and reflects a reliable application of those methods to the case. Daubert v. Merrell Dow Pharmaceuticals, Inc. (1993) and Kumho Tire Co. v. Carmichael (1999) direct courts to consider testing, error rates, peer review, and acceptance; the research summarized above bears directly on those factors for opinions about post-pubertal subjects. Military Rules of Evidence 701 and 702 parallel the federal rules, and Military Rule of Evidence 1102 generally incorporates amendments to the Federal Rules 18 months after their effective date. The December 1, 2023 amendment to Federal Rule 702 accordingly became part of Military Rule of Evidence 702 in 2025: the proponent of expert testimony in a court-martial must now show the military judge that it is more likely than not that each admissibility requirement is met, so gaps in an expert’s basis or in the application of the method are questions of admissibility rather than of weight for the members (Joint Service Committee on Military Justice, 2025). In courts-martial, the Manual for Courts-Martial definition of child pornography reaches visual depictions of actual minors and obscene depictions of persons who appear to be minors, which affects what must be proved about age.

Arizona

Arizona requires proof that the person depicted was under 18 at the time the depiction was created, adapted, or modified, and, for dangerous-crimes-against-children sentencing under A.R.S. § 13-705, under 15. Section 13-3556 permits the trier of fact to infer that a participant is a minor if the depiction, through its title, text, or visual representation, depicts the participant as a minor; the inference is permissive, not mandatory. Since 2025, the definition of “minor” in § 13-3551 also includes a visual depiction indistinguishable from an actual minor, defined as a depiction that a viewer would reasonably conclude shows an actual minor. Arizona Rule of Evidence 702, amended effective January 1, 2012, follows the federal rule and Daubert, and Arizona Rule 701 governs lay opinion. The same reliability questions about Tanner staging and visual estimation therefore apply in Arizona courts, and the distinction between the under-15 and under-18 thresholds can make the precision of an age opinion directly relevant to sentencing.

Special Issues in Legal Practice

The principles described above apply with equal force to every party in a legal proceeding. The science operates the same way regardless of who is asking the questions, and the research summarized here does not by itself favor either side. What differs between the parties is not the science but the vantage point — the particular places where these issues tend to surface, and the particular errors each party is best positioned to guard against. The two summaries below identify those recurring issues. They are offered for orientation, not as litigation strategy; in every case the controlling question is what the evidence in that case will, and will not, support, and the expert’s role is to inform the trier of fact rather than to advocate for either side.

Special Issues for Prosecutors

The research supports the use of images depicting clearly prepubescent children, and early pubertal development, as evidence of minority, and most federal courts permit juries to assess age from such images without expert testimony. Identification of known victims, file names, chat context, and other non-visual evidence provide stronger proof of age than appearance, and they are unaffected by the limitations of visual estimation. Prosecutors are often best positioned to select charged images for which minority is clear and to support age with independent evidence where it is available.

The research counsels caution about expert opinions that assign an age, or a conclusion of under 18, to post-pubertal subjects based on Tanner staging or general appearance. The developer of the staging system warned against that use, and studies of experts viewing images of adults have found frequent misclassification of adults as minors. Opinions of this kind face substantial challenge under the amended Rule 702, and images of subjects who may be near 18 present a risk of acquittal or appellate reversal if they rest on appearance alone. Automated age estimates are best treated as investigative tools rather than trial evidence.

Special Issues for Defense

The research supports challenging opinions that a post-pubertal person is under 18 when they rest on Tanner staging or visual impression. The Rosenbloom and Tanner letter, the studies of expert misclassification of adults, and the systematic review provide a documented basis for questioning the reliability of such opinions, and Katz illustrates that courts may exclude images or testimony when the subjects are near the age of majority. Defense counsel are often best positioned to determine whether the persons in disputed images can be identified, whether the material came from commercial adult sources with age documentation, and whether a qualified expert can address the limits of visual estimation.

The research also counsels caution about the limits of these challenges. They do not apply to images of clearly prepubescent children, which, in the federal non-production cases sentenced in fiscal year 2019, made up nearly all cases (U.S. Sentencing Commission, 2021), and courts have repeatedly held that juries may evaluate such images themselves. The studies showing misclassification of adults as minors do not show that a particular disputed image depicts an adult; minors can also be misjudged as older, and the uncertainty runs in both directions. When non-visual evidence such as victim identification or chat content establishes age, the scientific limitations of visual estimation have little bearing on the outcome.

This paper is for educational purposes only and is not legal advice. The author is a psychologist, not an attorney; summaries of law are general and must be verified against current authority or with a licensed attorney.

References

Anda, F., Le-Khac, N.-A., & Scanlon, M. (2020). DeepUAge: Improving underage age estimation accuracy to aid CSEM investigation. Forensic Science International: Digital Investigation, 32, Article 300921. https://doi.org/10.1016/j.fsidi.2020.300921

Anda, F., Lillis, D., Le-Khac, N.-A., & Scanlon, M. (2018). Evaluating automated facial age estimation techniques for digital forensics. In 2018 IEEE Security and Privacy Workshops (pp. 129–139). IEEE. https://doi.org/10.1109/SPW.2018.00028

Cattaneo, C., Obertová, Z., Ratnayake, M., Marasciuolo, L., Tutkuviene, J., Poppa, P., Gibelli, D., Gabriel, P., & Ritz-Timme, S. (2012). Can facial proportions taken from images be of use for ageing in cases of suspected child pornography? A pilot study. International Journal of Legal Medicine, 126(1), 139–144. https://doi.org/10.1007/s00414-011-0564-7

Cattaneo, C., Ritz-Timme, S., Gabriel, P., Gibelli, D., Giudici, E., Poppa, P., Nohrden, D., Assmann, S., Schmitt, R., & Grandi, M. (2009). The difficult issue of age assessment on pedo-pornographic material. Forensic Science International, 183(1–3), e21–e24. https://doi.org/10.1016/j.forsciint.2008.09.005

Eckert-Lind, C., Busch, A. S., Petersen, J. H., Biro, F. M., Butler, G., Bräuner, E. V., & Juul, A. (2020). Worldwide secular trends in age at pubertal onset assessed by breast development among girls: A systematic review and meta-analysis. JAMA Pediatrics, 174(4), Article e195881. https://doi.org/10.1001/jamapediatrics.2019.5881

Joint Service Committee on Military Justice. (2025). Amendments to the Military Rules of Evidence since 1984 (G. E. Maggs, Ed.; 2025 ed.). https://jsc.defense.gov/Portals/99/MRE%20Amendments%202025.pdf

Kloess, J. A., Woodhams, J., Whittle, H., Grant, T., & Hamilton-Giachritsis, C. E. (2019). The challenges of identifying and classifying child sexual abuse material. Sexual Abuse, 31(2), 173–196. https://doi.org/10.1177/1079063217724768

Kutz, T. J. (1999). Tanner staging and pornography [Letter]. Pediatrics, 104(4), 995–996.

Oliván-Gonzalvo, G., Sánchez-Quiroz, P. L., & de la Parte-Serna, A. C. (2021). Use of Tanner stages to estimate chronological age in alleged child pornography cases: A systematic review. Andes Pediatrica, 92(3), 470–477. https://doi.org/10.32641/andespediatr.v92i3.3374

Ratnayake, M., Obertová, Z., Dose, M., Gabriel, P., Bröker, H. M., Brauckmann, M., Barkus, A., Rizgeliene, R., Tutkuviene, J., Ritz-Timme, S., Marasciuolo, L., Gibelli, D., & Cattaneo, C. (2014). The juvenile face as a suitable age indicator in child pornography cases: A pilot study on the reliability of automated and visual estimation approaches. International Journal of Legal Medicine, 128(5), 803–808. https://doi.org/10.1007/s00414-013-0875-y

Rosenbloom, A. L. (2013). Inaccuracy of age assessment from images of postpubescent subjects in cases of alleged child pornography. International Journal of Legal Medicine, 127(2), 467–471. https://doi.org/10.1007/s00414-012-0765-8

Rosenbloom, A. L., Rohrs, H. J., Haller, M. J., & Malasanos, T. H. (2012). Tanner stage 4 breast development in adults: Forensic implications. Pediatrics, 130(4), e978–e981. https://doi.org/10.1542/peds.2011-3122

Rosenbloom, A. L., & Tanner, J. M. (1998). Misuse of Tanner puberty stages to estimate chronologic age [Letter]. Pediatrics, 102(6), 1494. https://doi.org/10.1542/peds.102.6.1494

Stathopulu, E., Hulse, J. A., & Canning, D. (2003). Difficulties with age estimation of internet images of south-east Asian girls. Child Abuse Review, 12(1), 46–57. https://doi.org/10.1002/car.781

U.S. Sentencing Commission. (2021). Federal sentencing of child pornography: Non-production offenses. https://www.ussc.gov/research/research-reports/federal-sentencing-child-pornography-non-production-offenses

Legal Authorities

Ashcroft v. Free Speech Coalition, 535 U.S. 234 (2002).

Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993).

Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999).

United States v. X-Citement Video, Inc., 513 U.S. 64 (1994).

United States v. Broyles, 37 F.3d 1314 (8th Cir. 1994).

United States v. Katz, 178 F.3d 368 (5th Cir. 1999).

United States v. Kimler, 335 F.3d 1132 (10th Cir. 2003).

United States v. Nolan, 818 F.2d 1015 (1st Cir. 1987).

United States v. Riccardi, 405 F.3d 852 (10th Cir. 2005).

18 U.S.C. § 2256.

10 U.S.C. § 934 (Article 134, Uniform Code of Military Justice).

Manual for Courts-Martial, United States, pt. IV, ¶ 95 (2024 ed.).

Ariz. Rev. Stat. Ann. § 13-705.

Ariz. Rev. Stat. Ann. § 13-3551.

Ariz. Rev. Stat. Ann. § 13-3556.

Fed. R. Evid. 701.

Fed. R. Evid. 702.

Ariz. R. Evid. 701.

Ariz. R. Evid. 702.

Mil. R. Evid. 701.

Mil. R. Evid. 702.

Mil. R. Evid. 1102.

Related topic papers: Child Sexual Abuse Material: Federal, Arizona, and Military Law · Digital Evidence and What It Can Show About the User · Culpability and Sentencing Factors in CSAM Cases · Expert Testimony in Online Sexual Offense Cases

Related topics from Arizona Forensic Services: Child Abuse Allegations · Courts-Martial