Face shape, with the arithmetic on show

What a face shape category actually is

A shorthand for how four measurements relate to each other. That is the whole of it. There is no biological classification of faces into seven kinds, no anatomical boundary between a long oval and a short oblong, and no authority that settles how many categories exist. Different systems use six, seven or eight, and they disagree at the edges because the edges are conventions.

That does not make the idea useless. It is a genuinely handy way to answer practical questions — which frames sit well, which haircut balances what — because those questions really do depend on proportion. It just means the label is a summary, and the measurements underneath it are the part with substance.

Which is why this tool prints them. At a length ratio of 1.21 you are called oval and at 1.23 you are called oblong, and those are the same face. A tool that hides the number cannot tell you that you were on a boundary, so it hands you a verdict where the honest answer is "either, and here is how close it was".

The four measurements

Forehead width, taken at temple level, across the widest part of the upper face. Cheekbone width, the widest horizontal measurement anywhere on the face for most people. Jaw width, measured at the gonial angles — the corners of the jaw below and slightly forward of the ears, not at the chin. Face length, from the top of the forehead straight down to the base of the chin.

All four are scaled into millimetres using your own iris, whose horizontal diameter sits close to 11.7 mm in nearly every adult and — crucially — lies in the same plane as the features being measured, so it introduces no perspective error of its own.

Face length is the weakest number, and we say so

The face mesh gives a topmost point that represents the top of the forehead as the model reconstructs it. It is not reading your hairline, because it cannot see one reliably: hair covers it, styles change it, and hairlines recede over a lifetime.

So face length carries more uncertainty than the three widths do, and length is half of the ratio that separates round from oval from oblong. If your result lands near one of those boundaries, this is the figure most likely to be responsible, and the honest response is that you sit between two categories rather than in one.

How head turn distorts the answer

This is the single biggest source of wrong results in face shape tools, and almost none of them correct for it.

When you turn your head, horizontal measurements foreshorten by roughly the cosine of the angle while vertical ones barely change. Turn fifteen degrees and your cheekbone width reads about three percent narrower while your face length reads the same — so the length-to-width ratio inflates, and the tool concludes your face is longer than it is. Everyone drifts toward oblong.

This tool reads your head pose from the face model's transformation matrix, divides the foreshortening back out of the ratio, and reports the angle it measured. Past about six degrees it warns you, because a correction is not as good as a square photograph and you should know which you are relying on.

Why the thresholds are not the ones in styling guides

Every guide to face shapes will tell you an oval face is about one and a half times as long as it is wide, and that the forehead and jaw are "slightly" narrower than the cheekbones. Those numbers describe a tape measure held against a head: hairline to chin, over cheekbone width.

A face mesh measures something else. Its topmost point is a reconstructed top of the forehead rather than your hairline, and its widest points span the full width of the face rather than the cheekbones alone. That is a shorter length over a wider width, so the same face reads about 1.2 here where a tape would say 1.5.

The jaw figure makes the problem concrete. Published anthropometric norms put jaw width at roughly 79% of cheekbone width in men and 77% in women, and faces measured here come out at 80% — close agreement, and a good sign the landmarks are sound. But it means a jaw at 80% of face width is the normal jaw. Any rule that treats "narrower than 92%" as a narrow jaw fires on every human being alive.

So the thresholds below are expressed as deviations from what an ordinary face measures with these landmarks, not as the absolute ratios the guides quote. Taking the guide figures literally is what makes a tool announce the same two shapes over and over, and it is worth checking any tool you use against a few faces you can already describe.

The rules, in the order they are tested

Classification is a sequence of tests, and the first one to fire decides the shape. The order matters: a face can satisfy more than one rule, and the earlier rules describe more distinctive outlines.

A typical face measures a forehead at 86% of its width, a jaw at 80%, and a length of about 1.15× its width. Those are the reference points everything below is measured against.

1. Triangle — jaw more than 4 points wider than typical, and wider than the forehead.
2. Heart — jaw under 89% of the forehead and more than 3 points narrower than a typical jaw. Both conditions, because a jaw somewhat narrower than the forehead is what most faces have.
3. Diamond — forehead and jaw both more than 4 points below typical, so the face is genuinely narrow at both ends rather than merely narrower than its cheekbones, which every face is.
4. Oblong — length more than 1.22× the face width.
5. Square — a jaw that is not narrow and sits more than 4% fuller than an elliptical face outline predicts, with length under 1.2× the width.
6. Round — length under 1.12× the face width.
7. Oval — what remains when nothing else fires. Balanced proportions with a length between the short and long thresholds.

Oval being the fallback is not an accident of implementation. It is what the category means: a face without a distinguishing proportion. That is also why standard styling advice is usually written for it — there is nothing to correct for.

Square and round differ by the jawline, not the size

Both are short faces. What separates them is what the jawline does on its way to the chin: a square jaw stays wide and turns a corner, a round one curves away steadily. Measuring that turn is harder than it sounds, and the first two attempts were wrong in instructive ways.

The first required a jaw wider than typical. No face measured here ever reached it, so square could never be returned at all. If a category never comes back, that is a bug in the thresholds rather than a fact about faces.

The second used the angle at the chin between the two jaw corners, which sounds like a measure of angularity and is not. A shorter face puts those corners closer to the chin, which widens the angle mechanically — across real faces the correlation between chin angle and face length runs to about −0.92. Square therefore meant "short", which is what round already means, and the two categories collided.

What is used now compares the jaw to an ellipse. Take a smooth elliptical outline through the cheekbones and down to the chin, and at the height where the jaw corners actually sit it predicts a jaw width of about 80% of the face width. Divide the real jaw by that and an elliptical face scores 1.00, a jaw that stays wide scores above, and one that tapers away scores below. Because the height is normalised and both widths share the same denominator, the figure does not care how long the face is.

The zero point is not a number anyone picked. An ellipse predicts 0.800 at that position, and the mean jaw across every face measured here is 0.803 — so an ordinary face really does score 1.00, and the scale means something before any threshold is applied to it. How far above 1.00 a jaw has to sit before the face reads square is still a judgement, and it is the number most likely to move.

The seven shapes

Why your result might differ from another tool's

Three reasons, in descending order of how often they are responsible.

Different thresholds. Every tool picks its own cut-offs and most do not publish them. If one calls a face long above 1.5 and another above 1.6, they will disagree about a large band of perfectly ordinary faces. Ours are printed above, and they are conventions rather than findings.

Different landmarks. "Jaw width" can be measured at the gonial angles, lower down toward the chin, or at the widest point of the lower face, and those give different numbers on the same photograph. Same for forehead width, which can be taken at the temples or higher up.

Head position. As above, an uncorrected head turn shifts the ratio enough to change the category. If you photograph yourself twice and get two answers, this is usually why.

Getting a reading you can trust

• Face the camera square on, chin level, looking at the lens rather than the screen.
• Pull hair back off the forehead and away from the jaw. Hair over the temples narrows the forehead measurement; hair over the jaw widens it.
• Use even, frontal light. Shadow along the jaw moves where the model places the jawline.
• Stand further back and zoom in rather than holding a phone at arm's length. A wide-angle lens close to the face enlarges whatever is nearest it and distorts the proportions.
• Keep a neutral expression. A wide smile lifts and widens the cheeks.
• Take two photos in separate sittings. If they disagree, you are near a boundary and the honest answer is that you sit between two categories.

Using the answer sensibly

The styling guidance attached to each shape is convention: accumulated practice about what tends to balance a given set of proportions. It is not a rule, it has no research behind it, and it is contradicted regularly by people who look excellent in exactly what they were told to avoid.

For eyewear specifically, it is also the weaker of two available approaches. Frame shape advice based on face shape is guesswork; frame size matched against measured face width and pupillary distance is arithmetic, and it is what actually determines whether a pair of glasses sits properly. Our frame size finder does the second, and it will tell you more about what to buy than this page will.

This tool measures geometry in a photograph on your device and transmits nothing. Face shape categories are a styling convention rather than a medical or biological classification, and the guidance offered reflects conventional practice rather than evidence. It expresses no view on appearance.

Questions people ask

How is face shape actually determined?

From four measurements and two ratios. The four are forehead width at temple level, cheekbone width at the widest point, jaw width at the gonial angles below the ears, and face length from the top of the forehead to the chin. The ratios that do the sorting are length divided by cheekbone width, and how the jaw and forehead compare to the cheekbones. A fourth figure, how full the jawline is against an elliptical outline, separates square from round once the ratios have narrowed things down.

Why do different face shape tools give me different answers?

Because the categories have soft edges and every tool picks its own thresholds, usually without publishing them. A face at a length ratio of 1.21 and another at 1.23 are the same face, but one gets called oval and the other oblong. Tools also differ in where they measure from, and many use thresholds copied from styling guides that do not match what a face mesh actually measures — which is why so many of them return the same one or two shapes for everybody. This one prints every ratio and every threshold, and says explicitly when your result sat close to a boundary.

Does a turned head change the result?

Considerably, and it is the most common reason a result looks wrong. Widths foreshorten when you turn your head while face length barely changes, so a turned head inflates the length-to-width ratio and pushes the answer toward oblong. This tool measures your head turn and divides the correction back out, then tells you the angle so you can judge whether to retake.

Where does the tool measure my face length from?

From the top of the forehead as the face mesh reconstructs it, down to the chin. This is the least certain number on the page and we would rather say so. The model is inferring where your forehead ends rather than reading your hairline, and hairlines vary enormously and recede over time. If your result sits near the boundary between oval and oblong, face length is the figure most likely to be responsible.

How many face shapes are there really?

There is no authoritative number, because the categories are a styling convention rather than a biological classification. Most systems use six or seven. This tool uses seven — oval, round, square, oblong, heart, diamond and triangle — because that set covers the distinct combinations of the ratios without inventing categories that differ only in name.

Is one face shape better than another?

No, and any tool implying otherwise is selling something. The categories exist to answer practical questions about what balances a given set of proportions — which frames, which haircut. Oval is often described as ideal, which really means that standard advice happens to work for it without adjustment. That is a fact about the advice, not the face.