How the attractiveness test is scored

The free attractiveness test runs in your browser. Your photo and the points found on your face stay on your device; we only receive anonymous counts.

It combines four parts of facial geometry into a score from 0 to 100, then converts it to the 1–10 scale you see:

• 01Symmetry (35%): how closely your eyes, brows, nose, mouth, cheeks and jaw match left to right. Small head turns are corrected before measuring.
• 02Facial thirds (25%): the balance between forehead, midface and lower face.
• 03Horizontal fifths (25%): how the eyes and the space around them divide the width of your face.
• 04Face length to width (15%): how long your face is compared with how wide it is.

The scale is set so that the middle adult face in a public research set of faces scores about 6.0. Scores run from 3.0 to 9.5: below 5.0 is "Below average", 5.0 to 6.4 "Around average", 6.5 to 7.4 "Above average", 7.5 to 8.4 "Well above average" and 8.5 or higher "Top tier".

Each attractiveness test online uses its own scale, so scores aren't comparable across sites.

The same photo file gets the same score under the same scoring version.

A different photo of the same face can score differently: light, angle, distance and expression all change what the camera sees. For the steadiest result, face the camera straight on in soft, even light, with a neutral expression.

After you pay, one compressed copy of your photo is sent to an AI model, together with your measurements. It looks at how the photo presents you: lighting and photo quality, expression, eye contact, hair, facial hair or makeup, and style. These are things you can change, so the report turns them into concrete fixes.

The report describes how your skin looks in the photo, never health or skin conditions; it never comments on ethnicity and never suggests surgery. Your photo is kept with your report for 12 months, and you can delete it any time.

The presentation scores come from an AI model, so a second run on the same photo can differ slightly. The geometry score does not.

We tested the scoring on AI-generated faces of fictional adults and on a public research set of photos of 102 adults who consented to research use (the Face Research Lab London Set).

• 01The same file scored the same every time.
• 02Photos turned more than about 5° to the side or tilted more than about 12° up or down are not scored, because head pose changes the measurements. Between 10° and 12° of up-or-down tilt, we show a result with a tip to retake at eye level for more comparable results. These limits are provisional photo checks, not scientific cutoffs.
• 03In our test, darker photos and darker skin tones were not wrongly rejected by the photo checks.
• 04We didn't see a clear difference in typical scores between men and women, or between ethnic groups. The sample is small, so this doesn't prove there is none.
• 05When we compared scores with a public set of rated faces, proportions explained only part of the picture. Expression, hair, lighting and styling tend to matter too, and those are things you can change.
• 06We don't claim a validated accuracy rate.
• 07Face shape is whichever of six typical shapes your five outline measurements sit closest to, and we also show the next closest. There is no labelled data to check the shapes against, so a face between two shapes could reasonably be called either.
• 08Our measurements match the published numbers The Face Analysis measured on the same research set with the same face-mesh model (face length to width, jaw to cheekbone width and eye tilt).

The method draws on published research on how people perceive faces. None of the authors is connected with this site or endorses it.

• 01Rhodes G (2006). The evolutionary psychology of facial beauty. Annual Review of Psychology, 57, 199–226.
• 02Todorov A, Porter JM (2014). Misleading first impressions: different for different facial images of the same person. Psychological Science, 25(7), 1404–1417.
• 03DeBruine L, Jones B (2017). Face Research Lab London Set. doi:10.6084/m9.figshare.5047666.v5 (CC BY 4.0). Used only for calibration; its photos never appear on this site.
• 04The Face Analysis (2026). Face Measurement Statistics 2026: 102 faces measured with MediaPipe. Updated 6 October 2026. thefaceanalysis.com (CC BY 4.0). Used to check our measurements.