why clothes don't fit.

Stand in any changing room and you can feel it before the mirror confirms it. The shoulders sit too far down the arm. The waist gapes at the back. The hem stops in a place that wasn't on the spec sheet for anyone's body. You take it off. You try the next size up. The shoulders fit. Now the waist is loose. You leave the shop, or you don't, and either way you've internalised something you weren't supposed to: that you are between sizes, that the clothes were made for someone else, that the fault is yours because you're standing here and the clothes are the clothes. This is the lived experience that everyone shares and no one prints on the label. Clothes don't fit. Not mostly. Not for outliers. They don't fit, in a structural sense, for almost everyone who isn't paid to wear them.

The numbers are quiet about it but consistent. The single largest reason people return clothing bought online is fit. Estimates vary by survey, but the studies converge on roughly half of all apparel returns being driven by size or fit issues, with one industry report putting the figure at 53 percent of online apparel return reasons.¹ The U.S. National Retail Federation valued returned merchandise in 2024 at $890 billion, around 16.9 percent of all retail sales, and retailers expect 2025 to come in close behind.² Clothing is the most-returned category online; a recent ICSC survey found that apparel ordered online has a return rate of 22 percent compared with 6.2 percent for the same items bought in a store.³ Sixty-three percent of consumers admit to deliberately buying multiple sizes of the same garment with a plan to return the ones that don't fit, a behaviour the industry has been forced to name: bracketing.⁴

None of that is a story about consumer pickiness. It's a story about a chart. The chart everyone is being measured against is older than most of the brands using it, was built from a wartime nutrition study with serious biases, was rejected by the standards body that commissioned it, was withdrawn by the government in 1983, and has been quietly rewritten by every retailer since. This essay is the story of that chart. What it is. Where it came from. Why the math behind it cannot, as a matter of geometry, fit most of the people who are graded against it. And what changes when you stop scaling a single template and start drafting from the body in front of you.

i. the wartime origin of the women's size chart

In the late 1930s the United States Department of Agriculture had a problem that wasn't, strictly, about clothes. Home-sewn garments were the norm. Pattern companies sold to women who drafted at the dining table. The patterns didn't fit because there was no agreed system of women's measurements. The Bureau of Human Nutrition and Home Economics, working with the Work Projects Administration, funded a study to fix it. Two researchers, Ruth O'Brien and William Shelton, set out to measure American women at a scale no one had attempted: 14,698 of them, across seven states, in 58 separate body dimensions each.⁵ The study, published in 1941 as USDA Miscellaneous Publication No. 454, was titled Women's Measurements for Garment and Pattern Construction. It is, despite everything that followed, one of the largest body-measurement surveys ever undertaken in the United States.

The problems are in the sample. Participants were paid volunteers, which selected, immediately, for women who needed the money. The study's criteria stipulated that participants had to be white; measurements taken from women of colour were excluded from the published statistics.⁶ The volunteers skewed young and skewed thin. Pregnant women were excluded. The study covered seven states out of forty-eight. It captured, in other words, a sample of one slice of one country at one moment, then offered the resulting averages as if they described the geometry of an entire half of the population.

O'Brien and Shelton themselves did not propose anything as crude as the size system that followed. Their data showed five to seven distinct body types within the sample, and their original proposal involved a three-part sizing convention, combining a numerical bust size with letters for height and lower-body girth, which yielded nine sizes per measurement. The Mail-Order Association of America, representing the catalogue retailers that depended on getting clothes to fit through the post, found this unworkable.⁷ The proposal was rejected.

What happened next mattered more than the study itself. The National Bureau of Standards re-analysed the data, dropped most of the dimensions, and published, in 1958, Commercial Standard 215-58: Body Measurements for the Sizing of Women's Patterns and Apparel. CS215-58 took the rich five-to-seven-shape spread of the original sample and collapsed it to a single hourglass figure, scaled by bust, with letter modifiers for height and girth.⁸ Sears and Montgomery Ward and the rest of the mail-order industry adopted it. Women in catalogues were now nominal sizes 8 through 38. The chart that produced those numbers was a single body type, sampled from a single demographic slice, normalised at a single moment, and applied to everybody who ordered a dress.

It did not work very well. Brands complained about the fit. The standard was made voluntary in 1970 and then withdrawn altogether by the U.S. Department of Commerce in 1983.⁹ ASTM International, the private standards body, published its own voluntary table in 1995, designated D5585 for misses' figure types and later D6960 for plus.¹⁰ Nothing has required any brand to follow them since.

ii. the math of grading, and why the size eight fits no one

A pattern is not a size. A pattern is a flat geometry drawn for a specific body. Before there were factories, every garment of any quality was drafted from the measurements of the person who was going to wear it. Industrial clothing made that economically impossible, so the industry built a workaround: pick one body, draft the pattern for that body, then scale the pattern up and down by fixed rules. The rules are called grading.

Here is how grading actually works in practice. A brand selects a fit model, typically a woman whose measurements approximate the brand's chosen size 8 — most commonly around five foot seven, with a bust-waist-hip of roughly 34, 26, 37 inches, although every brand picks a different fit model and every fit model is different.¹¹ The brand's pattern maker drafts a master pattern, called a sloper or a block, that fits this single woman. To produce a size 10, the pattern maker enlarges the size 8 master by a fixed grading increment, usually two inches at the bust, two at the waist, and two at the hip. To produce a size 12, the master is enlarged by another two inches at each landmark. And so on, in both directions.¹²

This is the load-bearing fiction. A real size 14 woman is not the fit model two inches inflated at three points. She has a different bust-to-waist ratio. Her shoulder slope is different. Her waist is at a different distance from her shoulder. Her arms are not necessarily two inches longer than the fit model's arms because her arms have no causal relationship to the fit model's bust. The grading rules assume that bodies are proportional enlargements of a single body. They are not.

Walk the math out a step. Suppose the master pattern's bust-to-waist ratio, set by the fit model, is 1.3 — a bust of 34 and a waist of 26. Now consider a woman whose actual bust-to-waist ratio is 1.45, which is well within ordinary anatomical variation. If her bust is 34 inches she should be a size 8 by bust, but her waist is around 23.4 inches. The size 8 pattern's waist is 26 inches. That is roughly an 11 percent gap at the waist, or about 2.6 inches of dead fabric, which a tailor would have to pinch out at every fitting. If she sizes down so the waist fits, the bust binds. There is no size in the chart for her. The chart isn't broken in her case; the chart simply assumes she doesn't exist.

This is the structural reason that bodies are blamed for fit failures. The label says size 14 and the size 14 doesn't fit because the size 14 isn't actually a fitted pattern for a size 14 woman. It is a size 8 pattern scaled by grading rules pushed three increments away from the fit model, then offered as if those increments described a real person.

SizeUSA, the 2003 anthropometric study that scanned roughly ten thousand American adults with full-body 3D scanners, found body proportions had moved a long way from any single template. Researchers analysing the dataset identified at minimum three dominant torso shape clusters — triangle, rectangle, and inverted triangle — with distributions that varied sharply by ethnicity. Black and white women clustered more towards the triangle shape; Asian women clustered towards rectangle; Hispanic women towards inverted triangle.¹³ The hourglass that the industry has spent seventy years scaling to and from is not the most common shape in any demographic group surveyed.

Add to that the obvious. Bodies change with age. A waist at 25 is not a waist at 55. A pattern that was correct for a fit model in her late twenties grades into a shape no one over fifty actually inhabits. Pattern grading does not adjust for the fact that the same nominal size has to dress different decades of life. It just stretches the master.

iii. vanity sizing, when the chart starts lying to you

The chart's first sin was that it was built from a narrow sample. Its second sin is that the numbers on it have been quietly moved over the years to flatter the customer.

The Wikipedia entry on vanity sizing assembles the standard comparison, which can be checked against the original ASTM tables. In 1958, by the Commercial Standard CS215-58, a women's size 8 had a 31-inch bust, a 23.5-inch waist, and a 32.5-inch hip. In the 2011 revision of ASTM D5585, a women's size 8 had a 35.25-inch bust, a 29.5-inch waist, and a 38.5-inch hip.¹⁴ Sit with that for a moment. A nominal size 8, in the same standards lineage, had grown five to six inches in every dimension across roughly fifty years. The body wearing the size 8 in 2011 was, in 1958 terms, a size 14 or a size 16.

This isn't an artefact of average bodies getting larger, although that has happened too. It is a deliberate marketing variable. The number on the label is now treated by most brands as an emotional signal, not a measurement. A 32-inch bust was a size 14 in the 1937 Sears catalogue. By 1967 the same 32-inch bust was a size 8. By 2011 it was a size 0.¹⁵ Nothing in the underlying anatomy of a 32-inch bust changed. The decision changed.

Economists have studied this. A working paper formally titled Economics of Vanity Sizing, published in the Journal of Economic Behavior & Organization in 2017, examined size charts of fifty-four American apparel retailers and found that sizes are inflated for women's brands at the moderately higher price points, and that the inflation is statistically correlated with brand positioning.¹⁶ Consumer psychology research in the Journal of Consumer Psychology has shown the obverse effect: when consumers receive a garment in a size larger than they expected, their appearance-related self-esteem drops and they engage in compensatory spending, a phenomenon the authors call "the flip side of vanity sizing."¹⁷ The brands are running a known psychology experiment on their customers, on which the customers cannot opt out, because the experiment is the label.

The result is sizing chaos. Try on a size 10 in five different brands and you will be a 10, a 12, a 6, a small, and an extra-small in succession. None of the brands are lying in the legal sense, because there is no legal definition of a size 10 in the United States or the United Kingdom for them to lie about. The standards exist; the standards aren't enforced; the standards have not been enforced for forty-three years.¹⁸ The chart is, functionally, whatever the brand says it is. The brand's incentives are to make the smaller-numbered sizes feel attainable and the larger-numbered sizes feel like they belong to other people. Tim Gunn, writing in the Washington Post in 2016, called this what it is: a disgrace. The average American woman, he reported, wears a size 16 to 18, and the industry refuses to make the clothes she would buy.¹⁹

iv. the made-to-measure middle ground

If the size chart is the problem, the obvious fix is to throw it out and have the garment cut to the body. This is what tailors have done for centuries, and what the made-to-measure industry now claims to do at internet scale. The claim is partially true and mostly not.

There is a useful three-way distinction here, which the industry tends to blur. The first category, ready-to-wear, is what almost everyone wears. One master pattern, drafted for one fit model, graded by fixed increments into a chart of nominal sizes. The customer picks a size and hopes. The second category, bespoke tailoring, in its strictest sense, is the opposite. A skilled cutter takes between twenty and thirty measurements of the customer, drafts a pattern from a body block constructed on those measurements, and constructs an intermediate garment called a toile or a muslin, cut from inexpensive fabric, to be tried on, marked, taken in, and re-cut before the final garment is touched. A serious Savile Row commission may involve three or four toile fittings before any wool is unrolled. The pattern that emerges does not exist for anyone else. It is drafted once, for this body, and kept in the cutter's archive.

Between those two sits the third category, the one that has expanded enormously online: made-to-measure. In its current commercial form, made-to-measure typically begins with a base pattern not unlike a ready-to-wear master, and modifies it by a small number of measurements, usually somewhere between five and a dozen, that the customer supplies. The base pattern is adjusted by the supplied numbers; the modifications might shorten the sleeve, raise the waistline, lengthen the rise. What it almost never does is throw the base pattern out and draft from scratch. The base pattern persists. The geometry of the master is still there, underneath, doing most of the work.

This is grading by another name. It is grading with extra parameters. It is more flexible than ready-to-wear, but it is not the bespoke tailor's approach. The toile, the in-person fitting, the iterative correction — the three things that make bespoke produce a garment that fits — are missing. What the customer is buying is a slightly more adjustable master pattern, sized to their stated measurements, and sewn from new fabric. If their measurements happen to be within the geometric range that the master can be stretched to cover without distortion, the result is acceptable. If their proportions sit outside that range, the master cannot be stretched far enough, and the result is grading-in-disguise: the same fundamental error that produced the ill-fitting size 14 off the rack, only this time the customer paid more and waited longer.

The reason the made-to-measure industry has not delivered on its promise has nothing to do with bad intent and everything to do with maths. The master pattern is a constraint. If your business model starts with a master and modifies it, you are still grading. The only way out of grading is to start with the body and draft to it. Doing that by hand requires a trained cutter and a toile and several appointments. Doing it at software cost requires something else.

There is a related thing the online made-to-measure industry rarely discusses, which is how the measurements are taken. Most services ask the customer to measure themselves with a tape measure, or supply a photograph and a height. Self-measurement with a tape measure has a known error envelope of two to four centimetres at the waist for an untrained user; that error then gets applied to the base pattern, and the result is a garment cut to a fictional set of measurements that combine the customer's body with their tape-measure technique. The fit improves on rack-bought clothing, often noticeably, because the customer's actual measurements are closer to the base than the chart's nominal size. But the pattern is still graded, the measurements are still approximate, and the toile is still missing. The promise the marketing makes — clothes cut to you — is half-kept.

That is also why most online made-to-measure brands publish their satisfaction rates and not their refit rates. Refit rates, when they are visible, tend to sit above twenty percent on the first garment, falling on the second after the brand has built a corrected pattern from the customer's feedback. Improving with iteration is exactly what one would expect from a system that is, structurally, still scaling a master.

v. what parametric pattern drafting actually does

Pattern drafting, as a craft, has been written down for almost as long as there have been books about it. Winifred Aldrich's Metric Pattern Cutting for Women's Wear, in continuous editions since 1975, gives explicit construction sequences for body blocks from raw measurements, with co-ordinate calculations that don't depend on a master pattern at all.²⁰ Helen Joseph-Armstrong's Patternmaking for Fashion Design, the textbook used in most American design schools, does the same thing for an American draft, building bodice, skirt, and sleeve blocks from individual measurements, then deriving styled garments from those blocks through dart manipulation, added fullness, and contouring.²¹ The knowledge has never been the bottleneck. Drafting by hand is slow. Doing it for one person costs hours. Doing it for ten thousand people costs ten thousand times the hours.

The technical work that escapes this is parametric pattern drafting: encoding the construction sequence in software so that the co-ordinates of every seam, dart, and notch are computed from the wearer's measurements, with no master pattern in the loop. Recent academic work in the International Journal of Fashion Design, Technology and Education describes the approach explicitly. In conventional CAD patternmaking the designer draws the pattern by clicking; in parametric drafting the construction is expressed as equations referencing a measurement table, and the geometry is generated.²² Change the table, and a new pattern is drafted. The geometry is downstream of the body, not a scaled version of someone else's body.

The remaining question, before the math can deliver, is how to get the measurements without sending the customer to a tailor. The current best answer is computer vision. Apple's TrueDepth camera, present on every iPhone since the iPhone X, projects an array of around thirty thousand infrared dots onto the subject and measures their displacement, producing a depth map accurate to roughly two millimetres at close range.²³ A short capture sequence, with the phone held at a fixed distance and the subject rotating, yields a sparse but reliable 3D landmark set covering shoulder, bust, underbust, waist, hip, arm length, rise, inseam, and the other dimensions a body block needs.

Combine the two and the loop closes. Capture twenty landmark measurements from the phone. Feed them into a parametric drafting engine built from Aldrich or Armstrong sequences. The engine outputs a flat pattern, in plotter-ready vectors, that has never existed before and will not exist again, because no one with exactly those measurements has been measured before or will be again. Two women who would both be a size 10 off the rack get materially different patterns, because their bust-to-waist-to-hip ratios differ, their shoulder slopes differ, their rises differ, and the engine sees those differences as inputs rather than as deviations from a master to be smoothed away.

This is not new in principle. It is what bespoke tailoring has always done. What is new is the cost. A century ago, getting clothes drafted from a body block required a trained cutter, several appointments, and a fabric budget. The labour cost was real and irreducible. The pattern engine moves the irreducible work from the cutter's hand to the software, which means the same geometry that used to be available only to people on Savile Row can be drafted by software for the cost of running the software. The garment still has to be sewn, which is a real labour cost. The pattern, which used to be the rate-limiting craft, becomes a file.

It is worth saying what this isn't, because the territory is crowded with overpromises. Parametric drafting does not abolish the difference between a good cutter and a mediocre one. The drafting rules in Aldrich and Armstrong assume a baseline body type and apply correction rules where the measurements deviate from that baseline. The corrections work well within ranges that the textbooks have been refined over decades to handle. They handle edge cases — very short rises, very sloped shoulders, unusual posture — less well. A skilled cutter looking at the customer in person will catch things the software won't, because the software is reading a depth map and not standing in the room. The honest version of the claim is that parametric drafting from camera-captured measurements produces, for most bodies, a fit that is closer to bespoke than to ready-to-wear, at a cost much closer to ready-to-wear than to bespoke. That is a large change. It is not a total one.

The other thing worth saying is that the output is a pattern, not a garment. The pattern still has to be cut, sewn, finished, pressed, and shipped. Those costs are real and dominate the price. The interesting design question, once the pattern problem is solved, is no longer "which size do you wear" but "what do you want made". The chart is what is being thrown out. The craft is not.

vi. what fit costs the world

It is worth being concrete about what the chart costs. Not because the cost is the moral argument — bodies deserve clothes that fit on their own terms, not because the alternative is wasteful — but because the cost is enormous, and almost entirely hidden from the people who pay it.

Start with returns. The U.S. National Retail Federation valued total returned merchandise in 2024 at approximately $890 billion, around 16.9 percent of all retail sales.² Apparel is the most-returned category, with online return rates averaging in the 22 to 30 percent range and some fast-fashion brands seeing returns above 38 percent.³ Across that volume, fit is consistently the leading cause; survey data places fit-driven returns at 38 to 53 percent of total apparel return reasons depending on the study.¹

A returned garment is rarely resold. Optoro, the reverse-logistics company that handles returns for major retailers, has reported repeatedly that returned inventory creates billions of pounds of landfill waste each year — five billion pounds annually in their earlier estimates, rising in their later reports to over nine billion pounds in the United States alone, accompanied by tens of millions of metric tons of carbon emissions.²⁴ Returns are dirty in a way the shopping experience is engineered to hide. The garment that didn't fit at home doesn't go back on the shelf. It goes to a sorter, then a liquidator, then, often, a hole in the ground or a container ship to a market that doesn't want it.

The scale of the broader textile waste problem dwarfs even returns. The United Nations Environment Programme estimates that 92 million tonnes of textile waste are generated each year globally, an amount equivalent to a refuse truck of clothing being incinerated or landfilled every second, with less than 15 percent of that fibre coming back into recycling streams.²⁵ Production roughly doubled between 2000 and 2015 while the average life of a single garment fell by 36 percent. Clothes are being made, sold, returned, abandoned, and disposed of at a pace the planet's waste systems cannot absorb, and a substantial share of that flow is driven by garments that didn't fit the people who ordered them.

The personal cost is quieter and is borne mostly in the body image of the people who own the clothes. Research published in the journal Body Image and elsewhere has shown that women with higher body dissatisfaction are also the women for whom clothes shopping is most negatively associated, that fitting rooms heighten body-focused anxiety, and that the experience of being too large or too small for a garment is processed by most people not as a fact about the garment but as a fact about themselves.²⁶ The label says size 12 and the size 12 doesn't fit; the customer concludes that her body has failed the chart, when the chart has, on its own terms, failed her. The systematic outcome of that misattribution is harm. It is not theoretical harm. It is harm large enough that a substantial body of clinical and consumer psychology literature has spent forty years trying to map it.

Add the money the customer spends on alterations, the time spent returning, the second outfit bought because the first didn't work, the third bought because the second didn't either. There is no single statistic that captures it because the costs are distributed and unrecorded. The shape of the loss is clear even when the dollar total is not.

And there is one more cost, harder to name. Most people, asked what size they wear, can answer immediately. Fewer can answer what their actual bust, waist, and hip measurements are. The number on the label has crowded out the measurement underneath it, and over time the substitution becomes the way the body is thought about: not in inches or centimetres, but in numbers that brands set. When the chart moves, the self-image moves with it, in a direction the customer did not choose. The chart was supposed to be a service. It became a vocabulary. The vocabulary now has more influence on how people see themselves than the bodies it was built to describe, and that influence is silent, persistent, and unaccounted for in any of the financial tallies above.

vii. the part that isn't broken

None of the history is anyone's fault, exactly. Ruth O'Brien and William Shelton were doing what nutrition researchers in 1939 could do. The Bureau of Standards was solving a real problem for catalogue retailers who had no way to know what to send their customers. The standards were eventually withdrawn because they didn't reflect reality, and were not replaced because there was no political appetite to require brands to follow a chart that none of them wanted to be held to. The grading rules were invented because factories needed a way to produce more than one garment from a single pattern. The marketing decisions to slide the numbers downward were the predictable consequence of a system in which no number on a label meant anything in particular and brands competed on emotion.

It is also true that something has changed in the last ten years that wasn't true before. The two pieces of equipment that bespoke tailoring assumed — a person who can take measurements, and a person who can draft a pattern from those measurements — are now plausibly replaceable by hardware most people already own and software that does what a craft did for a thousand years. The phone in the pocket has a depth sensor that, in 2017, would have been laboratory equipment. The drafting sequences in Aldrich and Armstrong are explicit enough that they can be written as code. The combination doesn't replace the tailor; the tailor is still better, and a serious bespoke garment will outclass anything a piece of software produces. The combination replaces the master pattern. It removes the assumption that one body has to stand in for every body. It makes the per-body pattern the default rather than the rare exception.

That, in the end, is the lever. Not a better size chart, because no size chart can fix a problem whose root is the existence of size charts. Not a better made-to-measure base pattern, because the base pattern is the constraint. Not better dressing-room signage about how a size 10 in one brand isn't a size 10 in another, because the customer should not have to think about that. The lever is to stop scaling from a template and start drafting from the body. The math has always been there. The body has always been there. What was missing was the cost of putting them together. That cost has changed.

The clothes most people wear were designed for someone they have never met, and the difference between that someone and themselves shows up every morning in front of the wardrobe. There is no virtue in pretending otherwise. There is, finally, no need to.