Methodology
How the AttractivenessTest score works
Every number on the result comes from the rules on this page. Nothing is hidden behind a model you cannot check, and the same code runs in your browser.
1. Your photo stays in your browser
The free tests load an open face model (MediaPipe Face Landmarker, Apache 2.0) into your browser. It places 478 points on your face. Our scoring code then measures those points. The photo, the points and the scores are never sent to us.
Before scoring, the photo has to pass a quality check: one face, looking straight ahead, eyes open, lips closed, fully in frame and evenly lit. A photo that fails is not scored; you get a hint on how to retake it instead.
2. What we measure
22 ratios and angles, each one free of image size and camera distance. You choose the men's or women's reference; we never guess it from the photo. The typical range is where most of the better-rated faces in our calibration set fall.
| Measurement | Part of | Typical, men | Typical, women |
|---|---|---|---|
| Middle to lower thirdNeoclassical canon: equal facial thirds; measured lower thirds tend to be slightly longer. | Harmony | 0.852–1.112 | 0.852–1.112 |
| Midface ratioPupil distance vs pupil-to-upper-lip height; a compact midface is preferred by convention. | Harmony | 0.936–1.147 | 0.936–1.147 |
| Eye spacing (rule of fifths)Rule of fifths: inner-eye distance equals one eye width. | Harmony | 1.055–1.348 | 1.055–1.348 |
| Face width in eye widthsRule of fifths: face width about five eye widths (mesh outline points sit slightly wider). | Harmony | 4.618–5.425 | 4.618–5.425 |
| Pupil distance to face widthTypical adult interpupillary distance relative to bizygomatic width. | Harmony | 0.429–0.489 | 0.429–0.489 |
| Mouth to nose widthMouth width about 1.5 to 1.6 times nose width. | Harmony | 1.481–1.763 | 1.481–1.763 |
| Nose width to inner-eye distanceNeoclassical canon: nose width equals the inner-eye distance. | Harmony | 0.833–1.043 | 0.833–1.043 |
| Lower to upper lip heightLower lip commonly about 1.5 times the upper lip height. | Harmony | 1.202–1.81 | 1.202–1.81 |
| Philtrum to chin heightPhiltrum about half the chin height. | Harmony | 0.35–0.515 | 0.35–0.515 |
| Canthal tiltA mild positive tilt (outer corner higher) is preferred by convention. | Harmony | 0.6–7.4° | 0.6–7.4° |
| Asymmetry indexMean mirrored deviation of paired points, as a share of face width; lower is more symmetric. | Harmony | 0.007–0.032 | 0.007–0.032 |
| Facial width-to-height ratioBizygomatic width over brow-to-upper-lip height; sex differences in research are small. | Dimorphism | 1.564–1.866 | 1.47–1.763 |
| Jaw to cheekbone widthBigonial over bizygomatic width; wider jaws are more typical of male faces. | Dimorphism | 0.788–0.836 | 0.758–0.799 |
| Chin height in lower thirdLip-to-menton height over subnasale-to-menton height; taller chins are more typical of male faces. | Dimorphism | 0.422–0.601 | 0.457–0.561 |
| Brow to eye distanceLower-brow-to-upper-lid gap over eye width; lower brows read as more masculine. | Dimorphism | 0.34–0.637 | 0.374–0.655 |
| Lower third share of face heightSubnasale-to-menton over glabella-to-menton height. | Dimorphism | 0.503–0.551 | 0.472–0.515 |
| Eye height to widthNarrower eye apertures read as more masculine, rounder as more feminine. | Dimorphism | 0.299–0.366 | 0.32–0.424 |
| Lip height to mouth widthTotal vermilion height over mouth width; fuller lips are more typical of female faces. | Dimorphism | 0.247–0.506 | 0.29–0.417 |
| Jaw angle (front view)Angle at the jaw corner between cheek outline and chin line; smaller reads as more defined. | Angularity | 126.8–138° | 126.8–138° |
| Chin to jaw widthChin width over bigonial width; a moderate taper reads as defined. | Angularity | 0.216–0.24 | 0.216–0.24 |
| Jawline curvatureMean jaw contour deviation from the gonion-to-chin line, as a share of face width. | Angularity | 0.04–0.051 | 0.04–0.051 |
| Face length to widthMesh top to menton over face width; mesh top sits below the hairline. | Shown only | — | — |
3. From measurements to scores
- Each of the 21 scored measurements has an ideal value. It scores 1 at the ideal and half that at one standard deviation away:
1 / (1 + (distance / spread)²). “Outside” on your result means the value lies beyond the typical range. - Harmony, Dimorphism and Angularity are weighted averages of their measurements, mapped to the 1–8 scale so that an average face in the calibration set lands near 4.8 and 8 means every measurement sits at its ideal.
- PSL = 0.35 × Harmony + 0.25 × Dimorphism + 0.2 × Angularity + 0.2 × Misc.
- Misc covers skin, hair, brows and grooming, which proportions cannot show. The free result therefore gives a range: the low end assumes Misc 3, the high end Misc 5.5.
4. How the ranges were set
The typical ranges, ideals and the 1–8 scale were fitted on the Face Research Lab London Set (DeBruine & Jones, 2017, CC BY 4.0): front photos of adults, each rated for attractiveness by 2,513 people. We measured 101 usable faces with the same code the site runs, and kept only summary numbers; no face from the set is shown or stored on this site. The dataset's authors are not involved with and do not endorse AttractivenessTest.
On those faces, our geometry score agrees with the human ratings only in part: a rank correlation of about 0.3 when checked on faces left out of the fitting. Proportions are one part of attractiveness; skin, hair, expression and style are others. Treat the score as a measurement of proportions, not a verdict.
5. Eye type
Canthal tilt is the angle from the inner to the outer eye corner, positive when the outer corner sits higher. Eye shape is eye height divided by eye width.
- Hunter: tilt of at least 4° and eye height to width of at most 0.31.
- Prey: tilt of 0° or less, or eye height to width of at least 0.38.
- Neutral: everything in between.
6. Limits
- Phone lenses held close make the nose and mid-face look wider. Hold the phone at arm's length.
- Light, head angle and expression shift the numbers. Compare two or three photos.
- The calibration faces were photographed in a studio in London; about two-thirds are white and all are adults aged 18 to 54, so the ranges may fit other groups less well.
- There is no global percentile: we do not have a sample of the world to rank you against.
- We do not give medical, cosmetic or surgical advice.
7. Versions
engine 0.2.0 · reference 0.2.0-frll · weights fixed-0.1.0
When we change a measurement, range or weight, the version changes too, so scores from different versions are not mixed up.