Two hand and wrist radiographs of a child lying side by side on a lightbox beside a sunlit window, each showing the small bones and open growth plates a radiologist reads to estimate skeletal maturity. A clinician's hand rests at the edge of the frame.

Growth science · Prediction

What bone age actually is — and why two hospitals can give different answers

GrowSense Growth Science · Educational, not medical advice

Every claim sourced to peer-reviewed research — see references below

You have two reports in front of you, for the same child, taken four months apart. One says bone age 8 years. The other says bone age 9 years. Same left hand. Same child. A whole year between them. So which hospital got it wrong?

Here’s the honest starting point: very often, neither did. Bone age is not a measurement in the way that height is a measurement. Nobody is counting something and getting the count wrong. Bone age is an estimate of skeletal maturity, produced by comparing your child against a reference model — and different hospitals may be using different models, different readers, and different reference populations. Two honest, competent answers can legitimately differ.

That sounds like bad news. It isn’t. It only becomes bad news if you treat a bone age as a verdict. Read correctly — as an estimate with a known margin, tracked over time — it is one of the most useful things you can know about your child’s growth. This guide explains what the number really is, the five reasons it moves between hospitals, how much movement is normal, and what to do when two reports disagree.

Bone age in one minute. Children grow taller because cartilage at the ends of their long bones is steadily converted into new bone — a process called endochondral ossification, driven by growth hormone, IGF-1, thyroid hormone and, in puberty, the sex hormones.[21][22] That cartilage layer is the growth plate. As a child matures, the growth plates progress through a predictable sequence — and eventually close, ending growth for good.[20][23]

Why the hand. The hand and wrist hold dozens of small bones and growth plates, all maturing on a broadly predictable timetable. A single low-dose radiograph of the left hand and wrist therefore captures a readable snapshot of the whole skeleton's progress.[4]

What the reader looks for. The size and shape of each bone, how much cartilage has turned to bone, and how the rounded bone ends relate to the shafts beside them.[20]

What the number means. A bone age of 9 doesn't mean the bones are nine years old. It means this skeleton looks like the reference skeleton of a typical nine-year-old. It is a comparison, not a count — and that single fact explains almost everything else in this article.

Want the biology? The growth plate as a living organ, how hormones drive it, and why estrogen closes it in both sexes are covered in When do children stop growing?

1. Two hospitals, one child, two numbers

Picture a nine-year-old girl. In March, a hospital near home reports her bone age as 8 years. In July, a larger children’s hospital reports 9 years. Her parents read that as a crisis: either she aged a year in four months, or somebody made a mistake.

What the parents concluded. "Her skeleton jumped 12 months in 4 months — she must be racing into puberty." Or: "One of these hospitals is incompetent."
What actually happened. The first hospital used Greulich–Pyle. The second used TW3. Those are different rulers with different reference populations — and a modest gap between them is expected, not alarming.[1][2][3] Her skeleton did not jump. The measuring stick changed.

Nothing about this child was unusual. What was unusual was expecting two different methods to produce the same number — and no one had told the parents the methods were different, because neither report said which method it used.

That is the real failure in this story. Not the radiologists. The report.

2. Bone age is a maturity estimate, not a diagnosis

Before we get to why readings differ, it’s worth being precise about what a bone age can and cannot tell you — because most parental alarm comes from asking it a question it was never built to answer.

A bone-age radiograph helps a clinician investigate short or unusually rapid growth, early or delayed puberty, possible endocrine conditions, chronic disease or undernutrition, how much growth potential remains, and whether a treatment is working.[4]

But it does not diagnose anything by itself. A delayed bone age is compatible with constitutional growth delay, endocrine disease, chronic illness or simple nutritional insufficiency. An advanced bone age may accompany early puberty, obesity or androgen exposure. The film supplies context; it never supplies the cause.

The rule that prevents most panic: a bone age is only interpretable next to something else — height velocity, pubertal stage, health history, and previous films. Alone, it is a number without a sentence around it.

3. Two different rulers: Greulich–Pyle and TW3

Almost every bone age you’ll encounter comes from one of two methods. They don’t just calculate differently — they think differently.

Greulich–Pyle (GP) works by pattern matching. The reader holds your child’s radiograph against an atlas of reference images and picks the one the whole hand most resembles. It’s fast — often a couple of minutes for an experienced reader — which is why busy hospitals prefer it. Its weakness is inherent to the approach: a child’s wrist may resemble one atlas age while the fingers resemble another, and the reader must decide which features to weigh most heavily. The atlas itself was also built from American children many decades ago, and researchers have since asked whether it still fits contemporary populations, with mixed answers depending on age, sex and ancestry.[9]

Tanner–Whitehouse 3 (TW3) works by scoring. Rather than matching the hand as one picture, it grades selected bones individually, assigns each a maturity stage, converts those to points, and totals them.[3] It’s more structured, less dependent on gestalt, and it exposes regional differences — one group of bones may be running ahead while another lags. The cost is time and training.

Here’s the part that matters to you:

GP and TW3 are not two routes to one true number. They use different reference systems and different reasoning. Large comparisons have repeatedly shown systematic differences between atlas-matching and bone-scoring.[1][2][3] A GP reading of 9.0 and a TW3 reading of 9.5 are not two hospitals contradicting each other. They are two instruments reporting in slightly different units.

Five things move the number between one hospital and the next. The first two live with the reader; the last three live in the method, the atlas and the film itself.

4. The reader weighs different bones

Skeletal maturation isn’t uniform across the hand. The carpals, radius, ulna, metacarpals and phalanges need not all match the same reference age. One radiologist may read the overall gestalt; another may lean on the distal radius or the phalanges. In young children, the arrival of new wrist ossification centres can dominate the estimate; through puberty, the shape and fusion of the bone ends take over.

This mosaic maturity is precisely why a whole hand can’t always be collapsed into an exact month — and why scoring systems like TW3 grade region by region instead of matching one picture.[3] Your child’s hand is not one age. It’s a committee, and the reader has to chair it.

5. Experience shifts the read

Studies with multiple readers generally find good to excellent overall reliability for GP, especially among trained readers.[5][6] Differences still occur, and training level contributes: one reliability study had paediatric radiologists and residents independently read the same films specifically to measure how experience moved the estimate.[7]

Other work reports strong agreement overall but warns that discrepancies cluster around puberty, when the skeleton changes fast and several atlas stages look defensible at once.[8]

That warning is worth holding on to, because of when it lands: the disagreement you’re staring at is most likely to appear in exactly the phase you’re most anxious about.

6. The methods may not be the same

This is the mundane one, and the most common.

One hospital uses GP. Another uses TW3. A third uses automated software trained to imitate one of them. If the report doesn’t name the method — and many don’t — you will compare two numbers believing they’re the same kind of thing when they aren’t. It’s a unit error, like comparing a temperature in Celsius to one in Fahrenheit without the little letter on the end.

7. The atlas may not match your child

A reference model is only as good as the children it was built from — and this matters enormously if your child isn’t from the population that built it.

A systematic review of 51 studies covering more than 20,000 children found that GP and TW3 were both precise — readers agreed with themselves and each other — yet their accuracy varied by ethnic group. In Asian and Arab children, both methods tended to overestimate bone age in adolescents approaching adulthood. In African youths, GP overestimated maturity where TW3 tracked more closely.[10]

A separate meta-analysis of the GP atlas alone found the mismatch runs in different directions at different ages: in Asian boys, GP read as delayed between ages six and nine, yet advanced by seventeen; in African girls, advanced.[11] A large Taiwanese comparison of GP against TW3 reached the same broad conclusion — population context changes how the result should be read.[3]

Precision and accuracy are not the same thing. A method can be highly reproducible — every reader agreeing — and still be consistently offset for your child's population. Reliable and correct are different claims.

None of this makes the methods useless. It means a bone age is an estimate produced by a specific reference model, and the further your child sits from that model’s original population, the more the number deserves a margin around it.

8. The image itself

The least glamorous explanation is often the right one. Hand positioning, exposure, image quality, and whether the whole hand and wrist are cleanly visible all affect what a reader can judge. Subtle maturity cues are simply easier to see on a technically good film.

And if the two radiographs were taken months apart, some of the difference may not be disagreement at all — it may be your child genuinely maturing in between.

9. So how much difference is normal?

There’s no single number that applies to every age, image and reader. Studies often report high correlation or good agreement — but high correlation does not mean two readers always assign the same age.

In practice, a difference of several months is common. Around periods of rapid change — early childhood, and puberty above all — gaps approaching six to twelve months can occur, because adjacent atlas standards are separated by broad maturity intervals and a real child often falls between two of them.

Two things decide whether a gap actually matters:

Your child’s age. A twelve-month difference means far more in a five-year-old than a fifteen-year-old, because it’s a much larger share of their remaining developmental time.

Whether the conclusion changes. A gap that moves the reading from clearly delayed to broadly age-appropriate is clinically meaningful. A gap that leaves the clinical picture unchanged is noise you can stop worrying about.

Reliability studies show GP can be reproducible,[5][8] but reproducibility never eliminates method-based and reader-based variation. The right conclusion isn’t “bone age is inaccurate.” It’s “bone age has a margin, and the margin should be stated.”

10. Why one reading can’t answer the real question

Here’s the deeper problem with the two reports on your table — deeper than which one is right. Neither one, alone, tells you what you actually want to know.

A single bone age describes skeletal maturity on one date. It says nothing about the speed maturation is travelling at. And speed is the thing that determines how much growing time is left.

Child A. Age 10, bone age 9. One year ago, bone age 8. Skeletal maturity advanced 12 months in 12 months — a normal, steady clock.
Child B. Age 10, bone age 9. One year ago, bone age 6½. Skeletal maturity advanced 30 months in 12 months — the clock is sprinting, and the growth window may be closing early.

Today’s number is identical. The futures are not remotely similar. Any single reading would have called these two children the same — and would have been wrong about one of them.

This is why serial readings answer questions a single film cannot: is maturation advancing normally, accelerating, or staying delayed? Is treatment maturing the skeleton faster than it’s adding height? Is the gap between chronological and skeletal age widening or narrowing?

The research agrees. When Tanner and colleagues built the TW Mark II prediction equations, they found that adding the previous year’s height gain or bone-age progression improved predictions at most ages over eight in girls and eleven in boys — and knowing last year’s height increment narrowed the interval for a ten-year-old boy from roughly ±8 cm to about ±4 cm.[15]

Read that again, because it’s the most important sentence here: one extra year of history halved the uncertainty. Not a better scanner. Not a cleverer algorithm. History.

This is not an argument for more X-rays. Repeat imaging must be clinically justified, and the interval must be long enough that real biological change exceeds measurement noise. In many situations, height velocity measured over six to twelve months tells you more than another film — and costs nothing. The case is for keeping and comparing the films your child already has.

For how much growing time actually remains, see When do children stop growing?

11. What bone age means for predicted height

Bone age feeds adult-height prediction because skeletal maturity implies how much runway is left. But a predicted height is an estimate with a margin — not an endpoint.

Traditional equations carry wide limits, especially in younger children and in those whose growth doesn’t resemble the population the equations were built on. In TW Mark II, 95% of predictions landed within about ±8 cm of true adult height for boys aged ten, tightening to roughly ±6 cm by fifteen.[15]

Automation improves consistency. Validation cohorts report root-mean-square errors near 3 cm — 3.3 cm for boys and 2.7 cm for girls in the Zurich longitudinal studies,[16] and 2.8 cm and 3.1 cm when the same model was tested on a French cohort.[17] Performance still varies by age, diagnosis and population.[18][19]

A prediction can legitimately move because of puberty timing, growth velocity, parental height, chronic illness, nutritional recovery, endocrine treatment, a change in skeletal maturation — or simply a different model. Which is why the responsible output is a range with its uncertainty stated, never a single confident centimetre.

The mechanics of prediction are covered in How tall will my child be?

12. What AI fixes — and what it doesn’t

Automated systems read the hand the same way every time, which removes one real source of variation. AI-assisted software has been shown to improve inter- and intra-observer agreement, with the largest gains among less-experienced readers.[12] Automated tools also show strong agreement with conventional reads in large clinical cohorts.[13] Deep-learning models can learn complex radiographic patterns directly,[14] while systems such as BoneXpert measure defined bones and compute maturity mathematically.[16]

But be clear about what this does and doesn’t solve. An algorithm inherits its training population, its target method, and its quality controls. If it was trained to reproduce Greulich–Pyle, it faithfully reproduces Greulich–Pyle’s population assumptions too — consistency is not the same as freedom from bias. It may also refuse abnormal, poorly positioned or technically unsuitable images.

So AI’s value isn’t declaring a hospital wrong. It’s providing a consistent second reading, region-by-region analysis, explicit quality checks, comparison against previous studies, honest uncertainty, and a maturity trend over time.

AI removes reader variation. It does not remove reference-model bias. An algorithm that is wrong in the same direction every single time is precise — and still wrong.

13. The report your child deserves

Most of the distress in this article traces back to one sentence:

What most reports say. Bone age: 9 years.

That sentence hides the method, the margin, the regional pattern and the history — everything you’d need to compare it with anything else. A more honest report looks like this:

Chronological age 9 y 10 m → Estimated bone age ~9 y → Method Greulich–Pyle → Difference ~10 months delayed → Confidence moderate → Regional pattern fingers and radius concordant; wrist slightly younger → Trend advanced 11 months over the previous 12 → Meaning mild delay; interpret with height velocity and puberty

Same child, same film, same radiologist. The difference is that the second version can be compared — with the last hospital’s reading, and with next year’s. It respects both the usefulness and the uncertainty of the test.

At minimum, every bone-age record you keep should carry: the method, the image date, the child’s chronological age, the reported bone age, the reader or software, and ideally an uncertainty range. Without those, two reports simply cannot be compared — and that, not radiological error, is why most parents end up frightened by a number.

14. What to do when two hospitals disagree

Practical, in order:

Find out which method each used. If one is GP and the other TW3, a gap of several months is expected and you may have no discrepancy at all. Ask; it’s often not written down.

Check the dates. Films months apart may show genuine maturation, not disagreement.

Ask whether the clinical conclusion changes. “Does this difference alter what we do next?” is the only question that matters. Usually the answer is no.

Bring the older films. A previous radiograph converts two competing snapshots into a trend — which, as Tanner showed, is worth more than either number alone.[15]

Don’t request a repeat X-ray to break the tie. A third opinion on the same maturity adds radiation and rarely adds information. If maturation speed is the question, the interval — not the tie-break — is what produces the answer.

Put height velocity beside it. Six to twelve months of accurate height measurements often tell you more than any of the films, at no risk.

One hospital said 8, another said 9. Should I get a third opinion?

Usually not. First ask which method each used — a GP/TW3 gap of several months is expected.[1][3] A third film adds radiation and typically doesn't resolve anything, because the disagreement isn't about the image, it's about the ruler.

My child's bone age is delayed. Is that good news or bad news?

By itself, neither. Delay is compatible with constitutional growth delay, endocrine disease, chronic illness or undernutrition.[4] It typically means more growing time may remain — but only the full picture, with height velocity and puberty, says what it means for your child.

We're Asian — is the atlas even right for my child?

A fair question, and the evidence says it deserves care. Both GP and TW3 have been shown to overestimate bone age in Asian and Arab adolescents nearing adulthood.[10] It doesn't invalidate the test; it means the result carries a margin, and your clinician should be reading it in context rather than to the exact month.

Is bone age the same as "growth plates open or closed"?

No. A delayed bone age suggests maturation is behind and more time may remain, but it isn't a direct measure of remaining centimetres. Different plates close at different times, and the outcome also depends on genetics, puberty and current velocity.[23]

Is an AI reading more accurate than my radiologist?

More consistent, which isn't the same thing. AI helps most where reader experience is limited.[12] But an AI trained on GP inherits GP's population assumptions — it can be reliably wrong in the same direction every time.

How often should bone age be repeated?

Only when it changes a decision, and never on a short interval — the gap must be long enough for real change to exceed noise. Height velocity over six to twelve months is often the more informative and completely risk-free measurement.

15. How this connects to the whole system

A bone age is one instrument reading one system, on one day. It cannot tell you whether your child slept badly all year, whether their protein intake collapsed during an illness, or whether puberty started three months ago. Yet all of those move the skeleton the film is measuring.

That’s why chasing the “correct” bone age is the wrong pursuit. The signal was never in the number — it’s in the relationship between chronological age, bone age, height percentile, height velocity, pubertal stage, health history, and change over time. Any one of those alone can mislead you. Together, they rarely do.

One number is a snapshot. A history is a signal.

GrowSense keeps every bone-age reading your child has ever had — across hospitals, methods and years — in one honest timeline, labelling what was measured versus estimated and which method produced each result. Not to tell you which hospital was right, but to show you the one thing no single X-ray can: the direction your child's growth is actually travelling.

Explore GrowSense

The parent takeaway

Bone age is not the age of the bones. It’s a model-based estimate of skeletal maturity — and like every estimate, it comes with a margin.

Two hospitals can produce different answers because they may use different methods, different reference populations, different readers, or different anatomical emphasis. A difference of several months rarely means poor care. Even a larger difference has to be judged against your child’s age, puberty, growth rate — and above all, whether it changes anything.

So when you’re holding two reports that don’t agree, the question to bring to the appointment isn’t “which number is correct?”

It’s “where is this child heading, and does either number change that?”

A single bone age is a snapshot. The trajectory is the signal.

References

Bone-age methods and how they compare

  1. Bull RK, Edwards PD, Kemp PM, Fry S, Hughes IA. Bone age assessment: a large scale comparison of the Greulich and Pyle, and Tanner and Whitehouse (TW2) methods. Arch Dis Child. 1999;81(2):172–173. PMID: 10490531.
  2. Milner GR, Levick RK, Kay R. Assessment of bone age: a comparison of the Greulich and Pyle, and the Tanner and Whitehouse methods. Clin Radiol. 1986;37(2):119–121. PMID: 3698492.
  3. Yuh YS, Chou TY, Tung TH. Bone age assessment: Large-scale comparison of Greulich-Pyle method and Tanner-Whitehouse 3 method for Taiwanese children. J Chin Med Assoc. 2023;86(2):246–253. PMID: 36652571.
  4. De Sanctis V, Di Maio S, Soliman AT, Raiola G, Elalaily R, Millimaggi G. Hand X-ray in pediatric endocrinology: Skeletal age assessment and beyond. Indian J Endocrinol Metab. 2014;18(Suppl 1):S63–S71. PMID: 25538880.

Reader reliability and observer variation

  1. Groell R, Lindbichler F, Riepl T, Gherra L, Roposch A, Fotter R. The reliability of bone age determination in central European children using the Greulich and Pyle method. Br J Radiol. 1999;72(857):461–464. PMID: 10505010.
  2. Paxton ML, Lamont AC, Stillwell AP. The reliability of the Greulich-Pyle method in bone age determination among Australian children. J Med Imaging Radiat Oncol. 2013;57(1):21–24. PMID: 23374549.
  3. Sariyilmaz K, Abali S, Ziroglu N, Cingoz T, Ozkunt O, Abali ZY, et al. Interdisciplinary and intraobserver reliability of the Greulich-Pyle method among Turkish children. J Pediatr Endocrinol Metab. 2023;36(12):1181–1185. PMID: 37844258.
  4. Faustino-da-Silva YV, Martinho DV, Coelho-E-Silva MJ, Valente-Dos-Santos J, Conde J, Oliveira TG, et al. Reproducibility and inter-observer agreement of Greulich-Pyle protocol to estimate skeletal age among female adolescent soccer players. BMC Pediatr. 2020;20(1):494. PMID: 33106161.
  5. Soudack M, Ben-Shlush A, Jacobson J, Raviv-Zilka L, Eshed I, Hamiel O. Bone age in the 21st century: is Greulich and Pyle's atlas accurate for Israeli children? Pediatr Radiol. 2012;42(3):343–348. PMID: 22237478.

Reference populations and ethnicity

  1. Martín Pérez SE, Martín Pérez IM, Vega González JM, Molina Suárez R, León Hernández C, Rodríguez Hernández F, et al. Precision and Accuracy of Radiological Bone Age Assessment in Children among Different Ethnic Groups: A Systematic Review. Diagnostics (Basel). 2023;13(19):3124. PMID: 37835867.
  2. Alshamrani K, Messina F, Offiah AC. Is the Greulich and Pyle atlas applicable to all ethnicities? A systematic review and meta-analysis. Eur Radiol. 2019;29(6):2910–2923. PMID: 30617474.

Automated and AI-assisted assessment

  1. Zhao K, Ma S, Sun Z, Liu X, Zhu Y, Xu Y, et al. Effect of AI-assisted software on inter- and intra-observer variability for the X-ray bone age assessment of preschool children. BMC Pediatr. 2022;22(1):644. PMID: 36348326.
  2. Bowden JJ, Bowden SA, Ruess L, Adler BH, Hu H, Krishnamurthy R, et al. Validation of automated bone age analysis from hand radiographs in a North American pediatric population. Pediatr Radiol. 2022;52(7):1347–1355. PMID: 35325266.
  3. Spampinato C, Palazzo S, Giordano D, Aldinucci M, Leonardi R. Deep learning for automated skeletal bone age assessment in X-ray images. Med Image Anal. 2017;36:41–51. PMID: 27816861.

Adult-height prediction

  1. Tanner JM, Landt KW, Cameron N, Carter BS, Patel J. Prediction of adult height from height and bone age in childhood. A new system of equations (TW Mark II) based on a sample including very tall and very short children. Arch Dis Child. 1983;58(10):767–776. PMID: 6639123.
  2. Thodberg HH, Jenni OG, Caflisch J, Ranke MB, Martin DD. Prediction of adult height based on automated determination of bone age. J Clin Endocrinol Metab. 2009;94(12):4868–4874. PMID: 19926715.
  3. Martin DD, Schittenhelm J, Thodberg HH. Validation of adult height prediction based on automated bone age determination in the Paris Longitudinal Study of healthy children. Pediatr Radiol. 2016;46(2):263–269. PMID: 26573823.
  4. Huang S, Su Z, Liu S, Chen J, Su Q, Su H, et al. Combined assisted bone age assessment and adult height prediction methods in Chinese girls with early puberty: analysis of three artificial intelligence systems. Pediatr Radiol. 2023;53(6):1108–1116. PMID: 36576515.
  5. Suh J, Heo J, Kim SJ, Park S, Jung MK, Choi HS, et al. Bone Age Estimation and Prediction of Final Adult Height Using Deep Learning. Yonsei Med J. 2023;64(11):679–686. PMID: 37880849.

Growth-plate biology

  1. Ağırdil Y. The growth plate: a physiologic overview. EFORT Open Rev. 2020;5(8):498–507. PMID: 32953135.
  2. Nilsson O, Marino R, De Luca F, Phillip M, Baron J. Endocrine regulation of the growth plate. Horm Res. 2005;64(4):157–165. PMID: 16205094.
  3. Lui JC, Nilsson O, Baron J. Recent research on the growth plate: Recent insights into the regulation of the growth plate. J Mol Endocrinol. 2014;53(1):T1–T9. PMID: 24740736.
  4. Cho JH, Jung HW, Shim KS. Growth plate closure and therapeutic interventions. Clin Exp Pediatr. 2024;67(11):553–559. PMID: 39463341.
← All Growth Science articles

This article is educational and does not provide medical diagnosis or treatment. Bone-age results must be interpreted by a qualified clinician alongside your child's growth rate, pubertal stage and health history. Imaging decisions, including whether and when to repeat a radiograph, should be made with your child's paediatrician or paediatric endocrinologist.