Methodology

What the puppy size calculator actually does, where the numbers come from, how well it performs when tested, and where it should not be trusted.

The growth model

Puppy growth is sigmoid: slow at first, fastest in the middle, then flattening toward an asymptote. We model the fraction of adult weight reached at a given age with a Gompertz curve, which is asymmetric in the way real growth is: the approach to full size is much slower than the climb toward it.

Two parameters control the curve, and both are functions of how large the dog will eventually be. The first is the age at which the dog reaches half its adult weight; the second is the growth rate constant. Making both depend continuously on adult weight matters, because the common alternative, sorting dogs into five size buckets, means a 51 lb dog and a 99 lb dog get an identical curve while a 49 lb dog gets a noticeably different one. Growth does not jump at round numbers.

Calibration

The curve parameters were fitted by least squares against percentage-of-adult-weight tables assembled from published breed growth charts spanning Chihuahua to Great Dane. Residuals after fitting, in percentage points of adult weight. The half-weight ages are quoted to one decimal deliberately: they are the model's own output for the representative weight in each row, not rounded category labels, and every percentage figure published elsewhere on this site is generated from the same function:

Size band	Representative adult weight	Age at 50%	Fit error (RMSE)
Toy	6 lb	14.6 weeks	3.2 pts
Small	16 lb	16.5 weeks	4.4 pts
Medium	35 lb	18.1 weeks	5.6 pts
Large	70 lb	19.5 weeks	1.7 pts
Giant	130 lb	20.7 weeks	3.9 pts

Tested end-to-end against fifteen weight-and-age points taken from published breed charts, the calculator recovers each chart's own stated adult weight with a mean absolute error of 6.9% and no meaningful directional bias (−1.3%). Worth keeping in perspective: different published charts for the same breed disagree with each other by 10–30% on adult weight, so this level of agreement is close to the ceiling the source data allows.

Independent cross-validation

Agreeing with the data you were fitted to is a weak test. The stronger one is agreeing with data you have never seen, so we checked the model against a source with no shared lineage with the breed charts it was calibrated on.

FEDIAF, the European pet food industry federation, publishes its own growth curves in its 2025 nutritional guidelines. They take a completely different mathematical form: a log-linear equation per mature-weight band, rather than a sigmoid function of adult weight. For the 27.5–47.5 kg band the equation is % of expected mature bodyweight = 36.96 × ln(age in weeks) − 56.18.

Run both for a 30 kg dog and compare across growth:

Age	FEDIAF	This model	Difference
12 weeks	35.7%	29.5%	−6.2
16 weeks	46.3%	40.7%	−5.6
20 weeks	54.5%	51.6%	−2.9
24 weeks	61.3%	61.5%	+0.2
32 weeks	71.9%	76.9%	+5.0
40 weeks	80.2%	86.7%	+6.6
52 weeks	89.9%	94.5%	+4.6

Across 12 to 52 weeks the mean absolute difference is 4.8 percentage points. For two models built independently, on different data, in different functional forms, on different continents, that is good agreement, and worth judging against the fact that published breed charts disagree with each other by 10–30% on the same breed.

It would be easy to quote only the 20-week row, where the two are under three points apart, and call it a validation. That would be selecting the flattering number. The disagreement is systematic rather than random: this model reads lower than FEDIAF before about 24 weeks and higher after it, crossing over in between.

The shape of that disagreement follows from the two functional forms, which is reassuring rather than alarming. A log-linear curve climbs fastest at the very start and then decays too slowly; a Gompertz curve has a genuine slow phase early and flattens toward an asymptote late. So the early gap and the late gap are the two forms doing what they structurally do.

Two further caveats, both of which cut in our favour and so deserve stating carefully. First, FEDIAF publishes one equation for the whole 27.5–47.5 kg band, while this model varies continuously: at 20 weeks it gives 52.1% for a 27.5 kg dog and 49.2% for a 47.5 kg one. Some of the difference at the band edges is therefore their binning rather than our error, which is the same argument for continuous parameters made further up this page. Second, a log-linear equation cannot asymptote: that band's equation passes 0% at 4.6 weeks and 100% at 68.4 weeks, so it is necessarily a local approximation valid inside a stated window, where a Gompertz curve approaches full size by construction. That is a difference in purpose as much as in quality. FEDIAF is sizing meals for a defined growth period, not projecting adult weight.

One honest limit on this check: we hold only the one band equation, so the comparison covers medium-to-large dogs and says nothing about toy or giant breeds.

Breed data

The calculator covers 268 breeds and varieties, including 59 designer crosses. Each carries a data-quality grade:

• Measured (74 breeds). Sex-specific adult weights from a study of 114,568 young adult dogs across 72 breeds. These are real population means rather than breed-standard ideals, which matters: breed standards describe show conformation, and pet dogs commonly sit outside them.
• Breed standard. AKC published weight ranges. Several entries in the source chart are stated maximums rather than ranges, or contain transcription errors; those were corrected against the breed standard and flagged.
• Designer crosses. No breed standard exists for these, so figures come from breeder-community datasets. Always shown as low confidence: a Goldendoodle's adult size depends heavily on which Poodle variety was used.

Where a breed's own sex split is unknown (which is the case for every AKC entry, because the published chart carries the same range in its male and female columns), males are estimated 20% heavier than females, holding the combined midpoint fixed so only the distribution between the sexes changes. That figure is not a guess: across the 74 breeds here carrying measured population data it is the median observed difference (interquartile range 17.6–23.0%), and males were heavier in every single one of the 72 breeds in that study without exception.

Two things we will not claim about it. First, the measured breeds run small through giant, with no toy breeds in the set, so applying 20% to a 5 lb dog is extrapolation, and only the direction of the difference is established there, not its size. Second, we tested a size-dependent version and rejected it: the measured medians are 19.7% for medium, 21.2% for large and 17.6% for giant, which is flat within noise on nine giant breeds. Fitting a curve to that would be fitting the noise.

The calorie calculator: two models, not blended

The dog calorie calculator is a separate tool with a separate method, and one decision inside it is worth stating plainly here. Every published source builds on the same base, resting energy requirement, which is 70 × (bodyweight in kg)0.75. They then part company, and not along the same axis.

• FEDIAF tiers adult dogs by activity and by age, and publishes no neuter factor.
• Merck tiers adult dogs by neuter status, and publishes no activity tiers.
• Measured pet dogs sit below both: Pedrinelli's 165 weight-stable animals averaged 86.1 against FEDIAF's headline 110.

A calculator that accepts activity and neuter status and returns a single figure has combined two models and added an assumption neither body published: that the neuter effect is multiplicative and independent of activity. We could find nothing testing that. So the tool evaluates each applicable method separately, attributes it, and reports the span. Reporting the disagreement is the method.

The headline figure is the lowest applicable method, following FEDIAF's own instruction to start from a lower calculated requirement and add as needed, given its warning that the recommendations "may overestimate energy needs by 10 to 60%". For an overweight dog the sums run on target weight rather than current weight, per AAHA 2021, because you feed the lean mass a dog has rather than the fat it is carrying. Full working and sources are in how many calories does my dog need.

What we deliberately do not model

Neuter timing

The mechanism is real: sex hormones trigger growth-plate closure, so early neutering delays closure and leaves long bones slightly longer. But the size effect is small. The largest analysis available, covering over three million dogs, found shifts of under one centile width and concluded that separate curves for neutered and intact dogs were unnecessary. Applying a numeric adjustment would imply precision the evidence does not support.

Paw size

The "big paws mean a big dog" heuristic has no quantitative form, and we could find no published study testing it in either direction. We leave it out rather than dress up an untested belief as a calculation. The evidence, including why the observation behind it is real even though the inference fails, is set out in do big paws mean a big dog?

Known limitations

• Giant breeds are the weakest case. The most rigorous published growth standards explicitly exclude dogs over 88 lb, because giant-breed growth varies too much between breeds to model with a single family of curves. Our giant estimates carry a wider range and an explicit warning for that reason.
• Under 12 weeks, treat output as indicative only. A puppy that young has not established the trajectory the projection relies on.
• The medium band fits about 5 points low. The published charts available for that weight range lean on brachycephalic breeds, which mature earlier than a same-weight herding breed. We report this rather than tune it away.
• Mixed breeds of unknown parentage are a wide guess. The underlying growth standards were built on purebred dogs and have never been validated on mixes.
• The cross-check covers one weight band. The FEDIAF comparison above runs on medium-to-large dogs, because that is the only band equation we hold. Toy and giant breeds have no external check at all.
• The validation set under-covers breed-standard entries. Ten of the fifteen test points are breeds with measured population data. Four take their adult weight from the AKC breed standard (the Beagle, Cavalier King Charles Spaniel, Chihuahua and Yorkshire Terrier), and one is a designer cross. So the 6.9% figure says much more about how the model performs on measured breeds than on breed-standard ones. The single worst point in the set is the Yorkshire Terrier, at +20.0%, and it is a breed-standard entry. We would rather say that than let one number stand for the whole dataset.
• No published error bars exist for this class of tool. No peer-reviewed validation study gives an accuracy figure for puppy-size prediction. Any calculator quoting a precise "±10% accurate" is making it up. Our 6.9% figure is agreement with published charts, which is a weaker claim than agreement with real outcomes.

Sources

• Salt C, Morris PJ, German AJ, et al. Growth standard charts for monitoring bodyweight in dogs of different sizes. PLoS ONE 2017;12(9):e0182064 — growth curve methodology, size categories, and the giant-breed exclusion.
• Salt C, et al. Comparison of growth patterns in healthy dogs and dogs in abnormal body condition using growth standards. PLoS ONE 2020;15(9):e0238521 — percentile crossing as an early signal of later weight problems.
• Andersson E, Emanuelson U, et al. Exploration of body weight in 115,000 young adult dogs of 72 breeds. Scientific Reports 2023;13:531 — measured sex-specific adult weights.
• Von Pfeil DJF, DeCamp CE. The epiphyseal plate: physiology, anatomy and trauma. Compendium — growth plate closure ages.
• FEDIAF Nutritional Guidelines for Complete and Complementary Pet Food for Cats and Dogs, September 2025 edition — independent growth curves (Table VII-8a), used above as an external check on this model.
• Purina Institute — size classes and skeletal maturity windows.
• American Kennel Club — breed weight chart and breed standards.
• Published breed growth charts used for calibration and testing: Pawlicy Advisor, Canine Journal, K9 Web, iHeartDogs, and owner-submitted breed datasets.