Understand your result

What Is a Good Face Symmetry Score?

A score of 97 needs a scale before it has a meaning. Here, it comes from the average reflected mismatch of 16 landmark pairs. Learn how to read the number, its rounding and the regional measurements without treating it as a population ranking.

In this guide
  1. Is 80, 90 or 95 a good face symmetry score?
  2. What does this checker actually measure?
  3. How the mismatch becomes a score
  4. Why a rounded 100 is not a perfectly symmetrical face
  5. Read the regions before averaging them
  6. What to do when your score changes
  7. Does a higher score mean a more attractive face?
  8. How accurate is an online facial symmetry score?
  9. Face symmetry score questions
  10. Try the formula with a known mismatch
  11. Sources and further reading
The numberA display index

The score expresses this tool’s particular calculation on a scale from 0 to 100.

The measurementMismatch / eye spacing

The percentage describes reflected landmark distances relative to a named reference.

The contextOne photo or capture

Pose, expression, perspective and estimated point positions can change the result.

Open the face symmetry checker Try the formula yourself

Is 80, 90 or 95 a good face symmetry score?

Those numbers become meaningful only after you know which method produced them. On this checker’s unrounded scale, 80 corresponds to 20% average reflected mismatch, 90 to 10%, and 95 to 5%, all relative to eye spacing. These are arithmetic examples, not “poor,” “normal” or “excellent” bands.

We have not established an average score for healthy people or a target you should reach. An unfamiliar-looking result is a reason to inspect the photo and landmarks, rather than a reason to assign yourself a label. Another app may use face width, a different midline, different points, different weights or a different score transform. Its 90 can mean something different from our 90.

A population percentile would need a suitable reference sample, controlled capture conditions and the same measurement method. Clinical thresholds would need a separate validation process. This tool has neither. The published methodology explains that distinction.

What does this checker actually measure?

The detector estimates facial points. We use the inner and outer corner midpoints to define two eye centers, level their connecting line, and place a vertical comparison line through their midpoint. For each selected pair, we reflect one point across that line and measure the straight-line distance to its counterpart. Smaller distances mean closer correspondence under this alignment.

From a reflected landmark distance to normalized mismatch Illustrative aligned coordinates, not a detected face. The eye centers are 100 pixels apart. A left point is reflected across their centerline. The corresponding right point is 3 pixels horizontally and 4 pixels vertically away from the reflection, a straight-line distance of 5 pixels. Dividing 5 by 100 gives 5 percent mismatch for this one pair. Reference line Eye spacing: 100 px Left point Reflected Right
Original geometric illustration, drawn to a consistent scale. Green points define the eye reference; the hollow blue point is the reflection and the brown point is its partner. The residual combines a 3 px horizontal gap and a 4 px vertical gap: √(3² + 4²) = 5 px. Dividing by 100 px eye spacing gives 5% mismatch for this one pair. The overall result averages all 16 pairs.

Eye spacing is the distance between those two corner-defined centers. It is not face width or a calibrated distance in millimeters. Normalizing by it makes the result independent of uniform image scaling: if every coordinate doubles, both the residual and the eye spacing double, leaving the ratio unchanged. It does not remove camera perspective, head turn or detector error.

The eye-based alignment also absorbs the relative height difference between the two eye centers. It cannot independently measure every eye-height difference. To inspect corner height relative to the original photograph, use the manual eye symmetry test, which has a different reference.

How the mismatch becomes a score

For each pair:
Pair mismatch = reflected distance ÷ eye spacing

Across all 16 pairs:
A = (pair mismatch₁ + … + pair mismatch₁₆) ÷ 16

Displayed measurement and index:
Mismatch percentage M = 100 × A
Score = max(0, 100 − M)

The score transform is a product convention. A one percentage-point increase in mismatch lowers the unrounded index by one point until it reaches the floor of zero. It does not count what percentage of a face is symmetrical.

A complete worked example

Suppose eye spacing is 100 px and the reflected distances of all 16 pairs add up to 48 px. The average distance is 48 ÷ 16 = 3 px. The mean mismatch is 3 ÷ 100 = 0.03, shown as 3.0% of eye spacing. The index is 100 − 3 = 97.

In the diagram above, one pair has a 5 px residual. If every one of the 16 pairs had that residual with the same 100 px eye spacing, the mean would be 5% and the index would be 95. One pair alone does not determine the whole result.

Examples for reading the scale

Illustrative inputs, calculated with this checker’s formula. These rows are not measured people or score bands.
Average pair residualEye spacingMismatchUnrounded indexHeadline
2.8 px100 px2.8%97.297
5 px100 px5.0%95.095
5 px200 px2.5%97.598
0.4 px100 px0.4%99.6100

The two 5 px examples show why the denominator matters: the same pixel gap is a smaller relative gap in a photo with larger eye spacing. They do not describe resizing the same photo, where the residual would change along with eye spacing.

Why a rounded 100 is not a perfectly symmetrical face

The headline rounds the index to a whole number; the main mismatch line rounds to one decimal place. In the last example, 99.6 rounds to 100 despite a nonzero residual. You cannot recover the exact mismatch from a rounded headline alone.

Even an unrounded index of 100 means only that the selected estimated pairs have zero residual under this alignment. The measurement does not cover every facial point, texture, depth or movement. It cannot establish a perfectly symmetrical anatomical face.

Read the regions before averaging them

Select a region in the checker to inspect its overlay and pair distances. The overall calculation gives each pair equal weight, so regions contribute different amounts according to how many pairs they contain.

How the current 16-pair method contributes to the overall average
RegionSelected pairsShare of the pair average
Eyes425%
Eyebrows318.75%
Nose212.5%
Mouth318.75%
Cheeks16.25%
Jaw318.75%

Consequently, averaging the six regional percentages equally will not generally reproduce the overall mismatch. Use the pair counts as weights. A larger circle on the mismatch map marks a larger selected-pair residual; it does not identify a medical problem or explain its cause. Bar lengths share a scale chosen for that result, so read the printed percentages when comparing different photos.

What to do when your score changes

Start with the input. A small turn, a smile, a nearer camera or an obscured landmark can change the projected points. Ward and colleagues’ photographic model illustrates how short camera distance changes apparent nasal proportions; it does not supply a symmetry-score correction. A photo can be sharp and still give a different geometric result.

  1. Check the photo. Use a frontal view at eye level, a relaxed expression, even light and visible features. Keep camera distance and camera settings consistent for a repeat.
  2. Inspect the detected points. Look at the region driving the mismatch. An estimated point that does not follow the visible feature reduces the usefulness of the result.
  3. Compare the raw measurements. In the two-photo comparison, 3% in photo A and 4% in photo B is a difference of +1 percentage point of eye spacing. It does not prove that your face changed.
  4. Record the conditions and method. If saving an export, keep the capture context with it. The export records the method version; a future method change may alter the numbers.

How to interpret the five-frame camera scan

The scan reports the median index across five accepted frames and the observed mismatch range. Regional details come from the frame nearest the median mismatch. For a hypothetical set of 2.6%, 2.8%, 3.0%, 3.2% and 3.4%, the median mismatch is 3.0%, the median index is 97 and the observed range is 2.6–3.4%. That range describes those frames; it is not a confidence interval, anatomical accuracy bound or proof of repeatability on another day.

For the causes in more detail, read why symmetry scores change and how selfie distance changes perspective.

Does a higher score mean a more attractive face?

This score has not been validated as an attractiveness predictor. Research on perceived attractiveness uses its own images, raters and methods. In a 2025 study by Lee and colleagues, symmetry did not independently predict attractiveness ratings for the 100 white adult face images studied. That result has a defined sample and does not settle every question about perception or any individual’s appearance.

Facial proportions ask a different question again: they compare the sizes of defined distances rather than left-to-right reflected points. A high symmetry number and a different proportion number can coexist without contradiction. Explore the face proportion calculator or our guide to symmetry and attractiveness research for those distinct questions.

How accurate is an online facial symmetry score?

There are three separate things to check: whether the arithmetic is implemented correctly, whether estimated landmarks match the photograph, and whether the result measures the real-world quantity you care about. Passing an equation test establishes the first; it does not establish the other two.

Lim and colleagues’ 2022 comparison of 2D photographs with direct measurements found that reliability and accuracy differed by facial dimension. That study supports checking a method for its intended measurements; it did not test this checker. Verhoeven and colleagues’ 2016 study compared landmark and surface registration methods on 3D images, a different measurement setup from this browser tool.

Our index has implementation tests and a disclosed method, but no human accuracy study, clinical validation or representative population reference dataset. Adding more points or a more complex equation would still require validation. MediaPipe’s documentation explains the landmark detector; it does not validate this site’s scoring scale.

Face symmetry score questions

Is a 90% face symmetry score bad?

First check what that website calls a percentage. An unrounded index of 90 here corresponds to 10% mean reflected mismatch relative to eye spacing. We have no validated good-or-bad cutoff. Inspect the input and points before interpreting the number.

What is the average face symmetry score?

We have not measured a representative population with this method, so an average is unavailable. Figures reported by another app do not establish an average for this checker. A credible reference needs its sample, capture protocol and formula disclosed.

Can a perfectly mirrored photo get 100?

A mirrored composite duplicates pixels from one side, but the detector still estimates landmark positions. A result depends on those estimates and the alignment. A rounded 100 is possible with a small nonzero residual and does not confirm anatomical perfection.

Why does a visibly different feature barely affect my score?

The index averages selected pairs. A local difference may contribute little to the overall mean, may fall outside the selected points, or may be absorbed partly by the alignment. Inspect the region and pair table instead of relying on the headline.

Can I confirm accuracy by getting similar scores on several sites?

Agreement alone does not establish accuracy. Sites can share a detector or similar assumptions while using different denominators and transforms. Look for documented validation against an appropriate reference for the actual measurement.

Should I aim to improve the score?

Use it to understand a photo and the method. There is no validated target to pursue. The index cannot choose a treatment, measure your worth or tell you how others perceive you. For a photo experiment, keep conditions consistent and record what you deliberately changed.

Explore the idea

Try the formula with a known mismatch

Change an illustrative residual to see exactly how this tool’s index is produced. This is arithmetic, not a population benchmark.

Chosen mismatch3.0%
Tool’s index97.0 / 100

100 − 3.0 = 97.0. This does not mean 97% of the face is symmetrical.

Why can’t I compare this with another website’s number?

The points, midline, weights and final transform can differ. Compare a score only with measurements made by the same method under matched conditions. Even then, a difference between photos does not prove a physical change.

Sources and further reading

The equation and interpretation describe this site’s current published method. The sources below support detector behavior, photography and research context; none validates this product’s score or supplies a “good” range.

Privacy settings

Photo analysis works without analytics. You can optionally allow Google Analytics to measure basic page visits. We do not send photos, filenames, face measurements or scores to analytics.

Analytics is off unless you allow it.

Your choice is stored in this browser. Turning analytics off after allowing it reloads this page; save any wanted result first. Read the privacy policy.

Check my face