How Accurate Are Growth Charts at Predicting Height?

Growth charts give a useful ballpark for where a child’s height is headed, but they are far from crystal balls. When used to project adult stature during childhood, the typical prediction error runs about 4 to 5 centimeters in either direction from age four through the start of puberty, and that margin widens considerably once the pubertal growth spurt enters the picture. The accuracy depends heavily on the child’s age, how close they are to puberty, and whether the chart was built from a population that actually resembles the child being measured.

What Growth Charts Actually Tell You

A growth chart does not predict your child’s adult height the way a weather model forecasts tomorrow’s temperature. It shows where a child’s current height falls relative to other children of the same age and sex. A five-year-old boy on the 75th percentile is taller than roughly three-quarters of boys his age. The implicit prediction is that he will stay near that percentile and reach an adult height somewhere around the 75th percentile for grown men. That assumption works reasonably well in many cases, but it rests on the idea that a child’s growth trajectory will continue to track along the same curve, which is not guaranteed.

Two major chart systems are in wide use. The World Health Organization charts describe how healthy breastfed children grow under optimal conditions and are considered a prescriptive standard. The CDC charts, by contrast, are a descriptive reference built from how American children actually grew during specific survey years. The WHO sample included only breastfed infants and was international and prospective, while the CDC sample reflected a heavier and somewhat shorter population, particularly in infancy.¹ In the United States, the CDC recommends using the WHO charts for children under two and the CDC charts from age two onward.² Which chart system is used can subtly shift where a child appears to fall, particularly in the first couple of years of life.

The Typical Margin of Error

If you take a child’s current height, look up their percentile, and assume they will land on the same percentile as an adult, you get a prediction with a standard error of roughly 4 to 5 centimeters for children between age four and the onset of puberty.³ That means about two-thirds of the time, the actual adult height will fall within 4 to 5 centimeters of the prediction, but one-third of the time it will be off by more than that. A prediction made at age six saying a boy will end up around 178 centimeters could easily land anywhere from 173 to 183 in practice.

More sophisticated methods narrow that window somewhat. A 2023 study comparing prediction approaches found that for boys between ages 8 and 14, the average error was about 3 centimeters using a method that compared individual growth curves to reference curves. For girls, the error was similar through age 12, then began dropping as they neared the end of growth.⁴ Machine-learning models trained on large cohorts have achieved average prediction errors of about 4 centimeters, with a correlation of around 0.87 between predicted and observed adult height. But these models still systematically overpredict height for shorter children and underpredict for taller ones.⁵

Why Puberty Scrambles Everything

The biggest single source of prediction error is the timing of puberty. Children who hit their growth spurt early shoot up the chart temporarily and look like they are headed for a tall adult height. Children who are late to puberty fall behind their peers and look shorter than they will eventually be. Because the percentile method assumes everyone grows at the same tempo, it overestimates adult height in early maturers and underestimates it in late maturers. During peak pubertal years, this mismatch inflates prediction error substantially. The percentile method’s average error was about 1.5 centimeters larger than a curve-comparison method for 13-year-old boys and about 1.7 centimeters larger for 12-year-old girls.⁴

An interesting finding from longitudinal studies is that pubertal timing affects total pubertal growth but not necessarily final adult height. Children who enter puberty later tend to have a smaller total growth spurt, but they have had more years of pre-pubertal growth to compensate. In girls, later breast development was associated with a smaller total pubertal gain in height, but adult height itself was not significantly correlated with the age of pubertal onset.⁶ The body has a remarkable ability to compensate, which is partly why the simple percentile snapshot during puberty is so misleading.

Mid-Parental Height and Target Height Formulas

The most widely used informal prediction is the mid-parental height formula: average the parents’ heights, then add about 6.5 centimeters for a boy or subtract 6.5 centimeters for a girl. This gives you what clinicians call the “target height.” It captures the genetic component reasonably well but has clear limitations. In a large analysis, the standard mid-parental height formula explained only about 36% of the variance in children’s heights, with heritability estimated at 74%. When researchers applied corrections for parental age and regression to the mean, the explained variance rose to 40% and prediction bias dropped dramatically, from 2.7 centimeters off target to just 0.14 centimeters.⁷

Even with those improvements, 60% of the variation in a child’s height comes from factors beyond what mid-parental height captures: nutrition, health during childhood, and the hundreds of genetic variants that do not neatly split between two parents. Moreover, the classic Tanner formula that many pediatricians still reference tends to underestimate final height in modern populations. A Korean study covering a decade of national survey data found that adults were, on average, about 4.9 centimeters taller than the Tanner formula predicted for men and 4.8 centimeters taller for women. Roughly 84% of the population ended up taller than their Tanner-based target.⁸ Some of that gap reflects secular trends in height, and some reflects formula limitations, but either way, the classic mid-parental height calculation should be treated as a rough guide rather than a reliable forecast.

Bone Age and Clinical Prediction Methods

When a pediatric endocrinologist wants a more precise prediction, they typically order a bone age X-ray of the left hand and wrist. The idea is that skeletal maturity tells you how much growing a child has left to do, which adds information beyond chronological age and current height. Several formal prediction methods use bone age, including the Bayley-Pinneau, Roche-Wainer-Thissen, and Tanner-Whitehouse models. Each has been around for decades and each has known biases.

A comparison of these methods in normally growing boys found that the Roche-Wainer-Thissen and Bayley-Pinneau models tended to overpredict adult height, while the Tanner-Whitehouse model underpredicted unless corrected for the maturation standards of the population being measured.⁹ The researchers noted that even under ideal conditions, with bone age readings done by the original developers of each method, “biologically significant prediction error remains.”

A major practical problem is that bone age readings vary between doctors. A study evaluating traditional height prediction found that manual bone age readings showed dramatic interobserver variability: up to 43% of predictions made by different readers differed by more than 5 centimeters when using the Bayley-Pinneau method.¹⁰ That means two equally qualified doctors reading the same X-ray could arrive at meaningfully different height predictions almost half the time. Automated bone age reading software has started to reduce this variability, producing more consistent readings and predictions that cluster closer to population means.

For families wanting a prediction without an X-ray, the Khamis-Roche method uses only the child’s current height, weight, and the heights of both parents. Its errors are only slightly larger than methods that include bone age, making it a practical option for healthy children without underlying growth conditions.¹¹ The original validation was limited to white American children, so applying it to other populations adds uncertainty.

Constitutional Delay and the Late Bloomer Problem

Constitutional delay of growth and puberty is one of the most common reasons parents seek height predictions, and it is also where predictions most often go wrong. These are the “late bloomers” who are short for their age and enter puberty later than average. Growth charts and even bone-age-based methods frequently overestimate their adult height, leading to reassurance that may not be warranted.

In a study of boys with constitutional delay, the average final height was about 165.7 centimeters, compared to a predicted adult height of 170.7 centimeters and a target height of 171.8 centimeters. That gap of roughly 5 centimeters between prediction and reality was statistically significant.¹² In other words, the standard clinical prediction told these boys they would grow several centimeters taller than they actually did.

Which prediction method works best for these children depends on how delayed their bone age is. For boys with a bone age delay of two years or less, the Bayley-Pinneau and Roche-Wainer-Thissen methods both provided reasonable predictions. For boys with more than two years of bone age delay, the Roche-Wainer-Thissen method was most accurate overall, while the Bayley-Pinneau method tended to overpredict.¹³ For girls with constitutional delay and short stature, no single prediction method proved superior, with errors ranging from underpredicting by about 2 centimeters to overpredicting by about 2 centimeters depending on the method used.¹⁴

Population Mismatch and Outdated Reference Data

Growth charts are only as representative as the population they were built from. A chart based on well-nourished Northern European children in the 1970s will not perfectly describe growth patterns in South Asian or Latin American children today. A systematic review of worldwide variation in child growth found that differences in head circumference between national and ethnic groups were large enough to risk misdiagnosing normal children as having abnormal head size when using a single international standard.¹⁵ Height shows similar variability. A review of international growth charts for children over five concluded that genetic, cultural, and socioeconomic differences across populations make it difficult to maintain a single international standard that works everywhere.¹⁶

Secular trends in height compound the problem. Children in most countries are taller today than children of the same age were 30 or 40 years ago, thanks to improved nutrition and health. Studies have found that growth charts based on data collected before 1990 show consistently shorter mean heights than more recent surveys.¹⁷ If your pediatrician’s office is using a chart based on decades-old data, your child may look shorter on paper than they truly are relative to their current peers. In practical terms, outdated British charts have been considered obsolete for individual growth assessment since the mid-1990s.¹⁸ Indian data similarly showed that 50th-percentile heights for boys were higher across all ages when newer data replaced 1989 references, and the 97th percentile at 18 years was nearly 2 centimeters taller.¹⁹

Growth Charts as Screening Tools

Pediatricians use growth charts not just to guess at adult height but to flag children who may have an underlying medical problem. A child who drops from the 50th percentile to the 10th over a year or two could have celiac disease, thyroid problems, growth hormone deficiency, or other conditions that slow growth. Once the condition is treated, many children experience catch-up growth and return toward their original trajectory.

The evidence for growth monitoring as a formal screening tool, though, is weaker than you might expect. A systematic review in The Lancet Diabetes & Endocrinology found that existing growth-monitoring practices rely on suboptimal methods, which can both delay the diagnosis of serious diseases and trigger unnecessary referrals. Of seven algorithms proposed to define abnormal growth, validation remained low and performance varied widely. The review also noted that WHO growth charts had poorer performance than some other charts for early detection of target conditions.²⁰ Finland stands out as a country that has integrated automated growth monitoring into children’s electronic health records, resulting in increased diagnoses of pathological short stature.²¹ Most other countries have not achieved that level of systematic follow-through.

A critical review went further, arguing that commonly used growth monitoring criteria are inaccurate predictors of inadequate growth later in childhood, and that simply collecting individual weight and height measurements does not reliably identify which children need support.²² This does not mean growth monitoring is useless. It means the current tools are blunt, and a single percentile reading at a single visit tells you much less than a series of measurements tracked over time. The pattern matters far more than the snapshot.

Genetic Scoring and Where Predictions Are Heading

One of the newer approaches to height prediction bypasses growth charts and bone age entirely, relying instead on DNA. Polygenic risk scores aggregate the effects of thousands of genetic variants associated with height. In one study using a large British cohort, a polygenic risk score identified children who would develop adult short stature with an accuracy approaching that of mid-parental height, achieving a discriminative power of 0.84 on a standard accuracy metric. The score could substitute for mid-parental height in existing prediction tools like the Khamis-Roche method without losing accuracy.²³

Genetic scores are also proving useful for understanding children with unexplained short stature. Among children referred for idiopathic short stature without a family history of shortness, a polygenic height score predicted heights substantially lower than mid-parental height would suggest, hinting at a genetic predisposition that conventional family-history-based methods miss entirely.²⁴ For clinical use, these tools are still mostly research-stage, but they point toward a future where height prediction draws on genomic data alongside traditional measurements.

On the imaging side, AI-based bone age reading is reducing the human variability problem. In a study comparing an AI program with two experienced human raters, the better rater achieved a mean absolute error of about 3.2 centimeters in adult height prediction, while the AI program came in at about 4.6 centimeters. A second human rater’s accuracy was similar to the AI’s, highlighting how much interobserver variation exists even among specialists.²⁵ For children with specific conditions like precocious puberty, combining AI bone age reading with clinical data like height percentile and mid-parental height has shown improved prediction consistency, even if the absolute accuracy gains over traditional methods remain modest.²⁶

Growth Hormone Deficiency and Treatment Predictions

For children with diagnosed growth hormone deficiency, prediction takes on a different character. These children are not just tracking along a low percentile; their growth has actively slowed due to a hormonal shortfall. Once treatment begins, the response in the first year is substantial. In a study of prepubertal children with idiopathic growth hormone deficiency, average height improved from about 2.5 standard deviations below the mean to about 1.7 standard deviations below after the first year, with a growth velocity of roughly 9 centimeters per year, which is well above the typical pre-treatment rate. By the second year, the velocity settled to about 7.4 centimeters. Prediction models incorporating variables like initial height, bone age, birth weight, growth hormone dose, and first-year treatment response explained about 77% of the variability in first-year height gains and 84% in the second year.²⁷

These treatment-specific models are far more tailored than general growth charts and are used by endocrinologists to set expectations and adjust dosing. They underscore a broader point: the more information you feed into a prediction, including hormonal status, bone maturity, treatment response, and parental heights, the tighter the forecast becomes. A growth chart percentile alone, without any of that context, is the roughest possible estimate.