Updated How we measure
What your result shows
Your result has three parts. The big number is your facial thirds: the share of your face height taken by the upper, middle and lower third, top to bottom. Under it, two bars draw your thirds and your fifths to scale, with ticks where an even split would put each edge, so you can see at a glance which part is longer or wider than the canon. Last come three width ratios, each next to the value MediaPipe’s average face measures with the same method.
None of these is a grade. They describe the shape of your face the way a tailor’s measurements describe a body: useful for choosing a haircut, glasses or a beard, and for understanding why your face reads the way it does, but not a verdict. The table below gives the average face’s values and the classical canon for each, so you have two reference points for every number.
| Measure | Mean face | Canon | What it is |
|---|---|---|---|
| Upper third | 32% | 33% | Hairline (estimated) to brows |
| Middle third | 33% | 33% | Brows to the base of the nose |
| Lower third | 35% | 33% | Base of the nose to the chin |
| Outer fifths | 21% each | 20% | Face edge to the outer eye corner |
| Eye fifths | 17% each | 20% | Corner to corner of each eye |
| Middle fifth | 24% | 20% | Between the inner eye corners |
| Eye spacing | 1.43 | 1.00 | Inner-corner gap ÷ one eye’s width |
| Nose ÷ mouth width | 0.73 | — | Nostril wings ÷ mouth corners |
| Jaw ÷ cheekbone width | 0.78 | — | Jaw corners ÷ cheekbones |
Notice that even the average face misses the canon on several rows. That is not a flaw in the average face; it is the main finding about the canons, covered below.
How the calculator measures
Google’s MediaPipe Face Landmarker places 478 points on your face in your browser. The calculator rotates them so the face looks straight at the camera, fits a line down the middle of the face, and measures every length either along that midline (the thirds) or across it (the fifths and widths). Because every result is a share or a ratio, the distance to the camera and the size of the photo don’t change it.
- Thirds. Measured along the midline between four levels: the estimated hairline, the point between the brows (glabella), the point where the nose meets the upper lip (subnasale), and the lowest point of the chin (menton). Each third is a share of the total, so they add up to 100%.
- Fifths. Measured across the face at eye level, between six points: the face outline on each side and the outer and inner corners of both eyes. Each fifth is a share of the face width at that level. They are listed left to right as the photo shows them; on an unmirrored photo, the first is your right side.
- Eye spacing. The gap between the inner eye corners divided by the average width of one eye.
- Nose ÷ mouth width. The distance between the outer edges of the nostril wings divided by the distance between the mouth corners.
- Jaw ÷ cheekbone width. The width across the jaw corners divided by the width across the cheekbones, the same ratio the face shape detector uses.
The scan measures eight frames of the front view and takes the median position of every point first, so one blink or twitch doesn’t move a ratio. A single photo is one frame. The full method is on the methodology page.
The facial thirds
The thirds are the oldest proportion rule for faces. Renaissance artists drew faces in three equal bands, and the rule still appears in drawing books, orthodontics and aesthetic medicine as a quick check of vertical balance. Each band covers a different part of the face:
- Upper third, hairline to brows: the forehead. Its length depends on where the hairline sits, which is why it is the least reliable of the three and the one hairstyle changes most.
- Middle third, brows to the base of the nose: the eyes, cheekbones and nose. It is mostly bone and changes little through adult life.
- Lower third, the base of the nose to the chin: the lips, teeth and chin. It is the most variable: the bite, the length of the chin, the teeth and ageing all change it.
How well do real faces fit? When Farkas and colleagues measured 153 young adults with callipers and tested nine neoclassical canons, the vertical canons fitted worst: their faces showed equality of no more than two of the parts at a time. The authors concluded that the canons do not represent average facial proportions and that treating them as a prescription for the ideal face needs testing.
The calculator compares your thirds with an even 33% each. A third within about 3.5 points of 33% counts as balanced in the result sentence; outside that, the sentence names the longest or shortest third and its share. On the average face the thirds read 32 / 33 / 35, a slightly long lower third.
The horizontal fifths
The rule of fifths says the face is five eyes wide at eye level. In the result’s second bar, each of your five widths is drawn as its share of the face width, with ticks at 20%, 40%, 60% and 80% where an even split would put the edges.
On the face mesh, even the average face doesn’t split evenly: the outer fifths read 21%, the eyes 17% and the gap between them 24%. Part of that is real (the gap between the eyes is often wider than one eye) and part comes from the mesh. Its eye-corner points are not the anatomical corners calliper studies measure from, and its face outline at eye level sits inside the ears. Compare your fifths with the average face’s before comparing them with 20%.
Calliper studies found the horizontal canons held more often than the vertical ones, but still in a minority of faces: in the Farkas study, the two that held most often (the space between the eyes equal to the nose width, and the nose a quarter of the face width) did so in 40% and 37% of young adults.
The width ratios
Eye spacing
The canon says one eye’s width between the eyes. On the mesh, the average face reads 1.43, for the reason above, and the result sentence calls eyes close-set below about 1.28 and wide-set above about 1.58. Research on spacing points to the average rather than the canon: Pallett, Link and Lee moved the eyes within the same faces and found they looked best with the eye centres about 46% of the face width apart, the spacing of an average face.
Nose ÷ mouth width
On the average face the nose is 73% as wide as the mouth. The result sentence calls the mouth wide for the nose below 0.64 and the nose wide for the mouth above 0.82. Nose width varies a great deal between populations: the international study by Farkas and colleagues, 1,470 adults in many ethnic groups, found nose heights and widths contrasted sharply with the North American reference in several groups. Mouth width changes with expression, which is why the scan asks for a neutral face.
Jaw ÷ cheekbone width
The average face’s jaw is 78% as wide as its cheekbones. A higher ratio means a squarer lower face, a lower one a face that tapers from the cheekbones to the chin. It is one of the four numbers behind the face shape detector, and it moves with body weight and with a beard that changes the outline of the jaw.
Is there an ideal set of proportions?
Plenty of sources list “ideal” facial ratios. When they have been tested, few hold up. Kiekens and colleagues measured 27 ideal ratios from the literature on frontal photos of 64 adolescents and compared them with ratings from 76 adults. Only two of the ratios were related to the ratings, and together with three profile angles they explained under 30% of the differences between faces.
What does predict ratings is closeness to the average. Composite faces made by averaging many real faces are rated more attractive than most of the faces in them, as Langlois and Roggman first showed in 1990, and the spacing study above found the best positions for the eyes and mouth were the average ones. That is how our attractiveness test uses proportions: even thirds, even fifths and an average eye spacing score well, and proportions carry 15% of the score. If you want to see how your face compares with the golden ratio specifically, the golden ratio face test shows that comparison, with the evidence against it.
What moves the numbers
- Head tilt. Tipping the chin down shortens the lower third in the image and lengthens the upper one; tipping it up does the opposite. The straightening corrects most of it; keep your chin level anyway.
- Camera distance. A camera close to the face enlarges the nose and the middle of the face relative to the edges. A study of short-distance photos showed the effect on nose width grows the closer the camera gets. Hold the phone at arm’s length.
- Expression. A smile widens the mouth and lifts the cheeks; raised brows lengthen the upper third.
- Hair, glasses and beards. Hair over the forehead moves the hairline estimate; glasses cover the eye corners; a beard can move the chin and jaw outline points.
- The model itself. MediaPipe’s model card reports a mean landmark error of about 2.6% of the distance between the eye centres over video and about 3.2% on a single image. Fusing eight frames brings the noise down; a single photo keeps it.
Reading a result: a worked example
Say your result reads thirds of 30 / 33 / 37, fifths of 22 / 16 / 25 / 17 / 20, an eye spacing of 1.52, a nose ÷ mouth width of 0.70 and a jaw ÷ cheekbone width of 0.84. Read it one line at a time:
- Thirds. The lower third, at 37%, is the longest and four points over an even 33%, so the sentence names it. The upper third is short at 30%, but it is the estimated one: check that no hair was covering the forehead before reading anything into it.
- Fifths. The middle fifth (25%) is wider than either eye (16% and 17%), as on most faces measured this way. The two outer fifths differ (22% against 20%). That can be a real difference between the sides, or a head turned slightly so the nearer side looks wider; a second scan facing straight tells the two apart, and the face symmetry test measures it directly.
- Eye spacing of 1.52 is above the average face’s 1.43 but inside the range the result treats as typical (up to about 1.58).
- Nose ÷ mouth at 0.70 is a little below the average face’s 0.73: a mouth slightly wide for the nose, well inside the typical range.
- Jaw ÷ cheekbone at 0.84 is above the average face’s 0.78: a squarer lower face, which the face shape detector would weigh towards square or oblong.
In one sentence: a slightly long lower face with a broad jaw and typical eye spacing. That is the kind of summary the numbers are good for, and it points straight at choices like a fuller hairstyle on top or a beard kept short at the chin.
Putting the ratios to use
The numbers are most useful for decisions about how you frame your face:
- A long upper third is often balanced by a fringe or by volume at the sides rather than on top.
- A long lower third is shortened visually by a fuller beard at the sides, or by hair with volume at the crown.
- Wide-set eyes suit frames with a darker or heavier bridge; close-set eyes suit a clear or thin bridge and slightly wider frames.
- A low jaw-to-cheekbone ratio (a tapering face) suits styles with width at jaw level; a high one (a square lower face) suits softer, rounder shapes.
The guides below cover each of these in depth, and the face symmetry test adds the left–right comparison the ratios leave out.
The free result and the full report
The free result is everything on this card: thirds, fifths and the three width ratios. The Full Face Report, $14.99 once, adds your attractiveness score broken into its six components, your profile from the side angles (jaw angle, chin and nose projection), symmetry by region and what would move your score, built on your device from the same scan.
Related reading
- Facial thirds and fifths
- The golden ratio in faces: what the evidence says
- Face shapes chart: the seven types
Last updated .
Sources
- Farkas et al., Vertical and horizontal proportions of the face in young adult North American Caucasians: revision of neoclassical canons, Plastic and Reconstructive Surgery 75(3), 328–338 (1985)
- Farkas et al., International anthropometric study of facial morphology in various ethnic groups/races, Journal of Craniofacial Surgery 16(4), 615–646 (2005)
- Pallett, Link & Lee, New “golden” ratios for facial beauty, Vision Research 50(2), 149–154 (2010)
- Kiekens et al., Facial esthetics in adolescents and its relationship to “ideal” ratios and angles, American Journal of Orthodontics and Dentofacial Orthopedics 133(2), 188.e1–188.e8 (2008)
- Langlois & Roggman, Attractive faces are only average, Psychological Science 1(2), 115–121 (1990)
- Ward et al., Nasal distortion in short-distance photographs: the selfie effect, JAMA Facial Plastic Surgery 20(4), 333–335 (2018)
- MediaPipe Face Mesh V2 model card, Google (2022)
- Kartynnik et al., Real-time facial surface geometry from monocular video on mobile GPUs, arXiv:1907.06724 (2019)