Golden ratio face test

Updated 15 October 2026How we measure

What your match means

The test measures four proportions of your face and compares each with its classical target: two against the golden ratio, 1.618, and two against the older drawing canons, which ask for equal parts. Each row gets a match from 0 to 100, where 100 means the ratio sits exactly on its target. Your overall match is the average of the four.

Read it next to one reference number. MediaPipe’s mean face, the average shape the face model is built around, matches 73%. Average faces are rated more attractive than most of the real faces they are made from, a finding that has held since Langlois and Roggman built composite faces in 1990. So the average face is not a golden-ratio face, and an attractive face does not need to be one either. A match of 60% or 85% describes how your proportions compare with an old convention; it is not a beauty score.

The table in your result shows, for each row:

• Yours: your ratio, to two decimals.
• Target: 1.618 for the golden-ratio rows, 1.00 for the canon rows.
• Mean face: the same ratio on MediaPipe’s average face, so you can see whether a gap to the target is yours or everyone’s.
• Match: how close your ratio sits to the target, from 0 to 100.

The sentence above the table names the row closest to its target and the one furthest from it. That is the most useful part of the result: it tells you which of your proportions differ most from the classical ideal, in which direction, and by how much.

How the test measures your face

Google’s MediaPipe Face Landmarker places 478 points on your face, in your browser. The test then rotates those points to face the camera straight on, so a head turned a few degrees doesn’t change the ratios, and measures four things. Every value is a ratio of two lengths, so it does not matter how far you sit from the camera or how big the photo is.

Two details matter. First, no face model can see your hairline, so the test estimates it: MediaPipe’s top point sits roughly halfway between your brows and your hairline, and the hairline is placed twice that distance above the brows. Hair over the forehead, a receding hairline or a high one all move the first row. Second, the mesh’s eye-corner points are not the anatomical corners that calliper studies measure from, which is why even the mean face reads 1.43 on the eye-spacing row rather than about 1. Compare your value with the mean face’s before reading much into that row.

Each row is scored on a ratio scale, so being 20% above the target and 20% below it cost the same:

In formula form: match = 100 × exp(−½ × (ln(ratio ÷ target) ÷ 0.25)²). The curve is flat near the target, so a ratio a few percent off still scores in the high 90s, and it falls fastest between 10% and 40% off. The full measurement method, including how the face is straightened and how the scan combines frames, is on the methodology page.

A worked example

Say your scan reads a face length ÷ width of 1.42, an eye spacing of 1.35, a mouth ÷ nose width of 1.45 and thirds of 30 / 33 / 37. The rows score:

• Length ÷ width: 1.42 is 0.88 of 1.618, a match of 87.
• Eye spacing: 1.35 against 1.00 is 1.35 times the target, a match of 49. The mean face reads 1.43 here, so this face’s eyes are actually a little closer-set than average.
• Mouth ÷ nose: 1.45 is 0.90 of 1.618, a match of 91.
• Thirds: the longest (37%) is 1.23 times the shortest (30%), a match of 70.

The overall match is the average, 74%, one point above the mean face. The result would name the mouth-to-nose row as closest to its target and eye spacing as furthest from it, and the table would show that the eye-spacing gap is shared by almost every face, while the long lower third is particular to this one.

What the golden ratio is

Divide a line into two parts so that the whole line relates to the longer part exactly as the longer part relates to the shorter. The ratio that does this is (1 + √5) ÷ 2, about 1.618, usually written φ (phi). Euclid described it around 300 BC as dividing a line in “extreme and mean ratio”. It turns up in pentagons and five-pointed stars, and the ratios of successive Fibonacci numbers (5 ÷ 3, 8 ÷ 5, 13 ÷ 8 …) close in on it.

The mathematics is real. The claims built on it are another matter. In a well-known review, the mathematician George Markowsky went through the most repeated examples, from the Parthenon to the proportions of the human body, and found that most rest on loose measurements, chosen points and generous rounding. With enough freedom over where to measure, a ratio near 1.6 can be found almost anywhere, because a great many everyday ratios fall between 1.3 and 1.8.

How the golden ratio got attached to faces

Two separate traditions get mixed together under “golden ratio face”. The older one is the set of neoclassical canons: rules that Renaissance artists used to draw faces, such as equal thirds from hairline to chin and a face five eyes wide. These ask for equal parts, a ratio of 1, not 1.618. The newer one came from aesthetic dentistry and surgery. In 1982 the orthodontist Robert Ricketts proposed that many facial proportions approach the “divine proportion”, and later Stephen Marquardt built a face mask out of golden-ratio shapes and promoted it as the ideal face.

That mask has been examined closely. A 2008 analysis in Aesthetic Plastic Surgery found that it best describes the proportions of masculinised white women, as seen in fashion models, that it is ill-suited to non-European faces, especially sub-Saharan African and East Asian ones, and that its proportions are inconsistent with what most people actually prefer. The author also showed that the usual way of checking a face against the mask was faulty.

This test keeps both traditions visible instead of blending them: two rows test the golden ratio, two test the canons, and the table labels which is which.

What the evidence says

If the golden ratio described attractive faces, faces closer to it should be rated more attractive. That has been tested directly, several times:

• Few golden proportions matter. Kiekens and colleagues had 76 adults rate photos of 64 adolescents, then measured 19 golden proportions as Ricketts defined them. Only 4 of the 19 were related to the ratings, and together they explained 16% of the differences between faces, which the authors judged too small to be of clinical importance.
• In 3D, no link at all. Rossetti and colleagues scanned 400 young adults in 3D and had a jury score them, then compared 10 facial ratios between very attractive and less attractive groups. Nine ratios showed no effect of attractiveness, and 7 of the 10 differed from 1.618 in both groups.
• The canons rarely hold. Farkas and colleagues measured 153 young adults with callipers and tested nine neoclassical canons. The two that held most often did so in only 40% and 37% of faces; the vertical canons held least. They concluded that the canons do not represent average facial proportions.
• The best ratios are the average ones. Pallett, Link and Lee moved the eyes and mouth within the same faces and asked which version looked best. The winners had eyes about 46% of the face width apart and an eye-to-mouth distance of about 36% of face length, the proportions of an average face, which they called the new “golden” ratios.

The pattern is consistent: the targets that track attractiveness are the population averages, plus symmetry and a few sex-typical traits, not 1.618. That is why our attractiveness test does not use the golden ratio anywhere in its formula, and why this page reports a match rather than a verdict.

Why a golden-ratio face would look long

The first row shows the gap most clearly. MediaPipe’s mean face is 1.37 times as long as it is wide when measured from the estimated hairline to the chin across the cheekbones. A face at 1.618 would be about 18% longer for the same width: in face-shape terms, a clearly oblong face. A face near the average, around 1.3 to 1.45 on this row, scores roughly 70 to 90 here, and a longer face scores higher simply because it is longer.

The mouth-to-nose row works the other way round for many people. Golden-ratio guides say the mouth should be 1.618 times the width of the nose. The mean face measures 1.38, so a wider mouth or a narrower nose moves you towards the target. Smiling widens the mouth, which is one reason the scan asks for a neutral face: otherwise your match would depend on your expression.

The four ratios, one by one

Face length ÷ width

Length runs along the midline from the estimated hairline to the lowest point of the chin; width is the distance between the widest face-outline points at cheekbone level. Hairstyle changes how long a face looks but not this measurement, except where hair covers the forehead and moves the estimate. A beard that changes the outline of the chin can move the chin point slightly.

Eye spacing ÷ eye width

The canon says the gap between the eyes equals the width of one eye, the familiar “five eyes across the face” rule. Because the mesh’s corner points are not the anatomical corners that calliper studies measure from, the mean face reads 1.43 here and nearly everyone reads above 1. A value close to the mean face’s means typical spacing; a value well above it means wide-set eyes for their width, well below it close-set.

Mouth ÷ nose width

Mouth width runs from corner to corner; nose width is the distance between the outer edges of the nostril wings. Both are soft tissue and both move: the mouth with expression, the nostrils slightly with breathing. The scan’s median over eight frames removes most of that.

Longest ÷ shortest third

The three thirds run from the estimated hairline to the brows, from the brows to the base of the nose, and from there to the chin. The canon says they are equal, a ratio of 1. The mean face reads 1.08, with a slightly longer lower third. The thirds are covered in depth by the facial ratios calculator, which draws them to scale on your result.

Getting a reading you can trust

Ratios are only as good as the points behind them. MediaPipe’s model card reports a mean landmark error of about 2.6% of the distance between the eye centres when it tracks a face over video, and about 3.2% on a single image. Two habits keep your result close to the real proportions:

• Hold the camera at arm’s length or further. A close camera enlarges whatever is nearest to it. A geometric study of short-distance photos showed that the closer the camera, the wider the nose looks relative to the rest of the face, which shifts the mouth-to-nose row, and the narrower the face looks, which shifts the length-to-width row.
• Use the scan rather than a photo. The scan measures eight frames and takes the median of every point, so a blink or a twitch doesn’t move a ratio. A photo is one frame.
• Face the camera, chin level, neutral expression. Tipping the chin down shortens the lower third in the image; smiling widens the mouth.
• Hair off your forehead, glasses off. Both hide points the ratios depend on.

What to look at instead

If you want to know how your face measures on the features studies do link to attractiveness, three other tools use the same scan:

• The face symmetry test compares 220 mirror pairs of points and shows which region differs most between your two sides.
• The facial ratios calculator shows your thirds and fifths to scale and the width ratios next to the average face.
• The attractiveness test combines symmetry, averageness, proportions, eyes, jaw and profile into one documented score.

The free result here is the match and the table. The Full Face Report, $14.99 once, adds the attractiveness score broken into its six components, your profile from the side angles, symmetry by region, thirds and fifths, and what would move your score, built on your device from the same scan.

Related reading

• The golden ratio in faces: what the evidence says
• Facial thirds and fifths
• Facial symmetry explained

Last updated 15 October 2026.