Face symmetry test

Updated 15 October 2026How we measure

What your symmetry score means

The score says how closely the two sides of your face mirror each other, from 0 to 100. A perfect mirror image scores 100. No real face does: every face has small differences between its left and right sides, so the useful question is how large yours are and where they sit.

Under the score, the result names the region where your two sides differ most and draws a bar for each of seven regions: forehead, brows, eyes, cheeks, nose, mouth and jawline. Each bar uses the same 0–100 scale as the overall score. The sentence under the number describes what the measurement found in plain words, for example which eyebrow sits higher or which side of the jaw is fuller.

For reference, the formula assumes that a typical face mismatches by about 3% of the distance between the outer eye corners, which scores about 74. That reference is a documented assumption, not yet a median of real scans, and it will be re-centred when real scans give a distribution.

How the test measures symmetry

The measurement runs in your browser on the 478 points that Google’s MediaPipe Face Landmarker places on your face. It takes five steps:

1. Straighten the face. The points are rotated so the face looks straight at the camera, which removes most of a head turn or tilt.
2. Find the midline. A straight line is fitted through 17 points that run down the middle of the face, from the forehead through the bridge and tip of the nose to the chin.
3. Mirror every pair. The mesh has 220 pairs of points that are exact mirror images on the average face, such as the two outer eye corners or the two mouth corners. Each point on your right side is reflected across the midline, and the test measures how far the reflection lands from its partner on your left side.
4. Remove what a head turn would explain. Points sit at different depths, so even a one- or two-degree turn that the straightening missed makes deep points like the jaw look lopsided. The test measures the exact pattern a tiny turn produces and subtracts the best-fitting amount of it.
5. Average and scale. The 220 distances are averaged and divided by the distance between your outer eye corners, so the size of the photo doesn’t matter. That average, d, becomes the score.

The score curve is 100 ÷ (1 + (d ÷ 0.05)²). It is flat near zero, so the small jitter every landmark model has costs little, and it halves at a mismatch of 5% of the eye-corner span:

The region bars use the same curve on each region’s own average. The overall score is not the average of the bars: it comes from the mismatch averaged over all 220 pairs, and because the curve is flat near zero, a face with one or two weaker regions can still score above most of its bars. The scan measures eight frames of the front view and takes the median position of every point before any of this runs; the figure after the ± in your result is how much the score moved between those frames. The full method is on the methodology page.

Why one photo can mislead you

Symmetry is the hardest measurement to take from a single photo, because the differences you are looking for are small, about the size of the landmark model’s own error. 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. A typical face’s real asymmetry is of the same order.

Two things in the test work against that. Averaging over 220 pairs means random errors on individual points largely cancel; and the scan’s median of eight frames removes most frame-to-frame jitter before anything is measured. The rest is down to how the photo was taken:

• Light from one side shades one half of the face, and the model places points less precisely in shadow. Face a window or a lamp.
• A turned or tilted head is mostly removed by the straightening and the head-turn step, but not completely. The scan only takes the front view when your head is close to straight.
• Expression. Almost nobody smiles symmetrically. Raised brows, a half smile or a squint show up directly in the brow, mouth and eye bars.
• A close camera exaggerates whatever is nearest the lens. Hold the phone at arm’s length.
• Hair and glasses cover the forehead and the eye corners. Hair back, glasses off.

If two scans in good light give scores within a few points of each other, the result is your face. If they differ by ten, the conditions are still moving it.

Reading a result: a worked example

Say your result reads 78, with the largest difference in the jawline, the jawline bar at 61 and every other bar above 75, and the sentence “Your jawline is fuller on your left side.” Before taking that at face value, run three checks:

1. Was your head turned? The jaw is the deepest part of the face outline, so a leftover turn shows up there first. Rescan facing the camera squarely.
2. Was the light even? A shadow along one side of the jaw hides its outline, and the model places those points less precisely.
3. Does a second scan agree? If the jawline leads again with a similar bar, the difference is in your face rather than in the conditions.

If it holds, it is an ordinary kind of asymmetry: a difference of a few millimetres in jaw width is common and can come from the bite or from chewing more on one side. A beard shaped slightly fuller on the narrower side, or a hair parting on the fuller side, balances it visually. The same reading applies to every region: check the conditions, rescan, and only then look at the direction the sentence gives.

Nobody’s face is symmetrical

Biologists separate two kinds of asymmetry. Fluctuating asymmetry is the small, random difference between sides that comes from development: two halves built by the same genes in a slightly different local environment never come out identical. Directional asymmetry is a difference that leans the same way across many people. Faces have both, and on top of that they move asymmetrically: emotions tend to show more strongly on the left side of the face, as an early study of half-face composites found.

Ordinary sources of a visible difference between the two sides include the shape and growth of the jaw, how the teeth meet, a nose set slightly off the midline, an eyelid crease that sits higher on one side, and the habits of expression that strengthen some muscles more than others. Most of these are a few millimetres. In normal faces they are small enough that observers mostly don’t pick them up: a study of portraits and their perfectly symmetrical composites found that the low level of asymmetry in normal faces did not change attractiveness judgments, except in portraits of older people, where symmetry seemed to read as a sign of youth.

Does symmetry make a face more attractive?

It is one of the most studied questions in facial attractiveness, and the answer is “yes, a little, with caveats”. The evidence for an effect:

• In one of the first studies to measure symmetry on computer images of faces, Grammer and Thornhill found that more symmetrical faces were rated more attractive.
• Perrett and colleagues changed only the shape of faces, keeping the skin texture fixed, and found that making the shape more symmetrical raised attractiveness ratings for both male and female faces.
• Identical twins share their genes but not every detail of their development. Comparing twin pairs, Mealey and colleagues found the more symmetrical twin was consistently rated more attractive, and the bigger the symmetry gap, the bigger the attractiveness gap.

The caveats:

• Scheib and colleagues found that symmetry predicted ratings even when people saw only half of a face, where symmetry can’t be seen, and that raters could not judge symmetry accurately. Symmetry seems to travel with other attractive features rather than acting alone.
• Swaddle and Cuthill made faces more symmetrical without changing average feature size and found the symmetrised faces were rated less attractive, possibly because they lost natural asymmetries and looked unemotional. Perrett’s later work suggests image-processing artefacts played a part.
• A meta-analysis by Van Dongen found clear signs of publication bias in this literature. Correcting for it cut effect sizes by 30–50%, and in the few studies with more than 100 participants the effect was close to zero.

That is why symmetry is one component of our attractiveness test, weighted at 20%, and why a perfect mirror is capped: it scores 74 on that component, not 100. A face that is perfectly symmetrical but otherwise typical would score about 5.9 instead of 5.0. A large mismatch pulls a score down; a very small one doesn’t carry it up on its own.

Mirror-half photos, and why you look different in pictures

A popular way to “test” symmetry online is to build two faces from one photo: the left half joined to its mirror image, and the right half joined to its mirror image. The two composites usually look like different people. That is a striking demonstration of how much the two halves differ, but it is not a measurement: each composite doubles one side’s width, so a face cut a few pixels off its true midline produces two very different results. The test above measures the differences instead of building faces from them.

Asymmetry is also why your face can look odd in photos. A mirror shows you your face flipped left to right, and that is the version you see every day. A camera shows the unflipped face that everyone else sees. In a classic study, people preferred their mirror image while their friends preferred their true image: each preferred the version they were used to. Our true mirror shows your camera feed un-flipped, live, if you want to see the version other people see.

Which side of your face is your good side?

Most people have a side they prefer in photos, and there is a pattern behind it. A study of painted and photographic portraits by Nicholls and colleagues found that the left side of the face is shown more often, with the head turned slightly to the sitter’s right. When they asked people to pose for a family portrait and show as much emotion as possible, people tended to turn their left cheek to the camera; when asked to pose as a scientist and hide emotion, they turned their right. Combined with the finding that expressions show more strongly on the left side of the face, that may be why the left side is the one people offer when they want to look warm.

For your own face, the region bars give a practical answer. If one side of your brows, eyes or mouth sits higher, that side reads differently in a three-quarter view. Take the same photo from both sides, in the same light, and compare: the side that looks more like you is your good side, whatever the portraits say.

Can you change your symmetry?

The bones that set most of the measurement don’t change without treatment. What changes quickly is how symmetrical you look and how symmetrical you measure:

• Expression and posture. A relaxed, neutral face held straight measures more symmetrical than a half smile with the head tipped.
• Grooming. A hair parting, the shape of the brows or a beard can balance a difference or make it more visible.
• Dental and jaw questions belong with a dentist or orthodontist, who can tell you whether a difference comes from the bite.

The free result and the full report

The free result is your score, the region that differs most, and the score for each region. The Full Face Report, $14.99 once, puts symmetry in context: your attractiveness score broken into its six components, symmetry by region with the direction of each difference, your profile from the side angles, and what would move your score. It is built on your device from the same scan.

Your golden ratio match and your facial ratios are separate tools; each runs its own 30-second scan on your device.

Related reading

• Facial symmetry explained
• Jawline and gonial angle
• The golden ratio in faces: what the evidence says

Last updated 15 October 2026.