Attractiveness test: what your photo can and can't show
A free attractiveness test that turns symmetry, averageness and proportion into a 0 to 100 facial harmony score, with the formula and error range shown, plus hair, glasses and neckline directions for your face shape. Measured on your device.
Front-facing photo, even light, hair off the forehead
Harmony score
0 to 100, with its weights and this photo's range shown.
Symmetry
Left against right for six feature groups, with a measurement floor.
Averageness
Distance from a disclosed reference set, the largest part of the score.
Proportion
The facial-thirds and golden-ratio gap, labelled as weak evidence.
A measurement is not a verdict on anyone's worth. Support: UK · US.
- A 0 to 100 facial harmony score with its full formula: averageness 50%, symmetry 30%, proportion 20%.
- Every score comes with the range this photo can support, and the three factors behind it are shown underneath.
- Points come from your position in a disclosed 43-face reference set, not from anyone rating you.
- Hair, glasses, neckline and collar directions for your face shape. Runs in your browser, no upload, no account.
How this attractiveness test works
If you typed in "am I attractive" or "how attractive am I" and want more than a guess, here is what this attractiveness test actually measures: three numbers this site's other tools already compute, reused here rather than invented for this page. The symmetry offset comes from the face symmetry test, the averageness distance is new but built from the same reference set, and the proportion gap comes from the golden ratio face test. Each is placed in the reference set and turned into points, 100 minus its percentile, and the facial harmony score is their weighted sum. The range beside it is the score re-read with symmetry anywhere inside this photo's own measurement floor. Averageness here is a geometric distance the research links to ratings averaged across many faces; it is not a measurement of your attractiveness.
What research links to facial attractiveness
A 2006 review in the Annual Review of Psychology names three candidates for biologically based standards of beauty: averageness, symmetry and sexual dimorphism. Its own words: "Averageness, symmetry, and sexual dimorphism are good candidates for biologically based standards of beauty. A critical review and meta-analyses indicate that all three are attractive in both male and female faces and across cultures." Those three, not a golden-ratio proportion, are what the field treats as measurable candidates.
Symmetry's own story turned out to be more specific than "symmetric faces are attractive." A 2007 study by Rhodes and colleagues measured symmetry, averageness, sexual dimorphism, perceived health and attractiveness across three large samples of faces, two of Western faces and one of Japanese faces. Symmetry correlated with attractiveness in every sample, but when perceived health was statistically controlled, only one of the six symmetry-attractiveness correlations remained significant. In the authors' words, "the appeal of symmetric faces was largely due to their healthy appearance, with most associations between symmetry and attractiveness eliminated when perceived health was controlled." Symmetry looks healthy, and a healthy appearance is doing most of the work symmetry gets credit for.
A 2025 study by Lee and colleagues in Scientific Reports tested symmetry, averageness and sexual dimorphism together in the same statistical model rather than one at a time, across a sample of rated faces. It reported that attractiveness was significantly predicted by averageness in both sexes and by femininity in female faces, but not by masculinity in male faces or by symmetry once all three were in the model together, concluding that averageness and femininity predict facial attractiveness judgments better than symmetry and masculinity do. That is one study, not a consensus, but it names the same two factors this page leads with.
| Factor | What research found | What this test measures | Limits |
|---|---|---|---|
| Averageness | Predicted attractiveness in both sexes even when tested alongside symmetry and dimorphism (Lee et al. 2025) | RMS distance from a disclosed reference set across six proportions | Reference set is 43 synthetic faces, not a population sample |
| Symmetry | Correlated with attractiveness in every sample, mostly through looking healthy rather than directly (Rhodes et al. 2007) | Left-right offset for six feature groups, with a measurement floor | A turned head or a smile changes it more than a real asymmetry would |
| Proportion (weak evidence) | Not named among the three candidates in the 2006 review (Rhodes 2006) | Gap from eight classical targets, including facial-thirds balance | No face in our set landed within 5% of every classical target |
How the score is built, and how far to trust it
thefacereport.com/tools/attractiveness combines symmetry, facial thirds, facial fifths and golden-ratio proximity into a score out of 100 with weights of 35, 25, 25 and 15 percent, which it calls its own calibration. pfpmaker.com/attractiveness-test prints a score out of 10 with an equivalent PSL rating. This page also prints one number, and publishes its weights the same way: averageness 50%, symmetry 30%, proportion 20%. The order follows the research above: averageness still predicted ratings with the other factors in the model, symmetry lost most of its effect once perceived health was controlled, and the golden ratio is not among the candidates at all. No study has validated these exact percentages, including ours.
How much should any landmark score be trusted? The MEBeauty benchmark ran 21 published facial beauty methods on one multi-ethnic dataset under one protocol. The three geometry-only methods, which measure landmark positions as this page does, reached a correlation of about 0.31 to 0.38 with the averaged human ratings. Deep models trained directly on those ratings reached about 0.76. That gap is why this is called a harmony score and shown with its three parts: it measures geometry well and attractiveness only loosely. The rated datasets those deep models learn from are licensed for non-commercial research only, which is one reason this page does not use one.
The words beside the number describe position, not looks: very close to the reference centre (80 and up), closer than most of the reference set (60 to 79), mid-range for the reference set (40 to 59), further out than most of the reference set (20 to 39), far from the reference centre (0 to 19). The score carries a range because it has to: re-reading symmetry anywhere inside a photo's own measurement floor moved the score by up to 24 points on the four example faces below, and a different photo of the same face can move it further.
Measured examples
Each card below is an AI-generated example portrait, not a real person, run through the same attractiveness test code the widget above runs. The default order is neutral; sorting by averageness is a choice you make, not the order the grid opens in, because no card here is more or less attractive than another.
A measurement is not a verdict on anyone's worth, including your own. If looking at numbers about your face is making you feel worse rather than curious, the BDD Foundation's email helpline (UK) or the International OCD Foundation's BDD program (US) are real places to talk to someone about it.
Why these attractiveness scores disagree
Search "rate my attractiveness" and run the same photo through several of the results, and you will get several different numbers, and none of them is necessarily wrong about its own arithmetic. Each site picks its own set of landmark points, its own weights for combining them, and its own reference set for a percentile, and none of that is usually disclosed. A face landing in the 80th percentile on one site and the 55th on another can be a fact about two different reference sets, not about the face.
The photo itself moves the numbers before any of that. This site's own face symmetry test corrects head rotation using the model's pose matrix and still finds that a smile, side lighting or hair over one temple changes which side measures larger. The golden ratio face test shows the same thing for proportion: the ratio depending on an estimated hairline missed its target by over 40 percent on the median face in our set, every time, in the same direction, which is the signature of a measurement problem rather than 43 unusual foreheads. An attractiveness test result changes when the photo changes, even when the face does not.
What to wear for your face shape
Stylists and opticians mostly apply one rule, contrast: add length to a round face, width to a long one, softness to a square jaw. The table is that convention, not research, and it marks the two shapes where the sources below disagree. After a scan, the card beside your score works out the hair and glasses directions from your own measurements instead of from the shape name.
| Face shape | Necklines | Shirt collars | Glasses |
|---|---|---|---|
| oval | Most necklines work; scoop, V and boat all keep the balance you already have. | Most collars work; a classic point or semi-spread is the safe default. | Square, Aviator, Browline |
| round | V-necks and open square necks add a vertical line; high crew necks repeat the roundness. | Point or button-down collars lengthen; wide spread collars add width. | Rectangular, Square, Browline |
| square | Scoop and sweetheart necklines soften a strong jaw.Not settled: One school says a square neckline echoes and flatters a square jaw instead. | Point or rounded club collars soften the angles. | Round, Oval, Cat-eye |
| oblong | Boat, crew and off-the-shoulder necklines add width; deep V-necks add more length. | Spread and cutaway collars add width; long point collars add length. | Oversized, Deep square, Aviator |
| heart | V-necks and sweetheart necklines balance a wider forehead against a narrower chin. | Semi-spread collars keep the jawline from looking narrower. | Oval, Round, Aviator |
| diamond | V-necks follow the narrow chin; boat necks widen the narrower forehead and jaw.Not settled: Sources split between a V-neck (echo the chin) and a boat neck (widen the top). | Semi-spread collars balance wide cheekbones. | Oval, Cat-eye, Browline |
Sources: All About Vision, frames by face shape (optician-reviewed) · Inside Out Style, necklines · Art of Manliness, shirt collars by face shape.
What no photo can measure
No attractiveness test, including this one, can see any of the following from a single photo.
Skin
Texture, tone evenness and complexion are not captured by landmark points at all; they need pixel-level analysis this page does not attempt.
Expression and movement
A still photo freezes one instant. How a face moves when it talks or smiles is part of how people read it, and no frame shows that.
Voice
Voice pitch and prosody carry their own, separate research on attraction, entirely outside what any photo can show.
Grooming and style
Hair, makeup, facial hair and clothing all change how a face reads, and all of them are choices rather than geometry.
Context
The same face reads differently in a passport photo, a wedding photo and a candid one. This page measures one photo, not a person across contexts.
Rater agreement
The 2007 study cited above averaged at least 24 raters per face and trait rather than trusting any one person's rating.
Attractiveness test questions
How attractive am I?
+
This attractiveness test gives you a facial harmony score from 0 to 100, and shows exactly how it was made. It measures three things from your photo: how far your proportions sit from a disclosed reference set (averageness), your left-right symmetry, and your gap from classical proportions. Each becomes points, and the score is their weighted sum: averageness 50%, symmetry 30%, proportion 20%. The score also comes with a range, because one photo can only be measured so precisely. It is a summary of geometry, not of attractiveness, which also depends on skin, expression, voice, grooming and who is looking.
Are AI attractiveness tests accurate?
+
Treat any of them, including this one, as measuring specific geometric properties rather than attractiveness itself. A 2025 study of averageness, symmetry and sexual dimorphism found that facial attractiveness was significantly predicted by averageness in male and female faces and femininity in female faces, but not by masculinity in male faces or by symmetry once the three were tested together. That is one study's finding, not a settled fact, but it argues against any tool that treats symmetry as the main driver of a combined score. A benchmark that ran 21 published methods on the same multi-ethnic dataset found geometry-only methods, the kind that measure landmark positions as this page does, agreed with average human ratings at a correlation of about 0.31 to 0.38, while deep models trained on those ratings reached about 0.76. So a landmark score tells you a lot about geometry and only a little about how people will rate a face.
Does a symmetrical face make you more attractive?
+
Symmetry does correlate with rated attractiveness in the research, but the more interesting finding is why. A 2007 study across three large samples of Western and Japanese faces found the appeal of symmetric faces was largely due to their healthy appearance: once perceived health was statistically controlled, only one of the six symmetry-attractiveness correlations remained significant. Symmetry looks healthy, and healthy looks attractive, more than symmetry itself is doing the work. A separate 2025 study testing symmetry alongside averageness and sexual dimorphism found symmetry did not significantly predict attractiveness at all once those other factors were in the model. This page reports your symmetry offset on its own, with a measurement floor, rather than folding it into anything.
Is the golden ratio the standard of beauty?
+
Not by the researchers who study facial attractiveness. The candidates a 2006 review names as biologically based standards of beauty are averageness, symmetry and sexual dimorphism. The golden ratio is not among them. This site's own golden ratio face test measured 43 AI-generated faces against eight classical proportions and found not one landed within five percent of every target. A large gap from 1.618 is the normal outcome, not a defect. This page reports that same proportion gap as weak evidence, clearly labelled, alongside the two factors the research treats as stronger candidates.
What is the attractiveness scale?
+
There is no single validated attractiveness scale in the research this page draws on. Consumer sites invent their own: pfpmaker.com/attractiveness-test prints a score out of 10 with an equivalent PSL rating and a percentile; thefacereport.com/tools/attractiveness combines symmetry, facial thirds, facial fifths and golden-ratio proximity into a score out of 100 using weights of 35, 25, 25 and 15 percent that it calls its own calibration, not a scientifically universal formula. No published study establishes those percentages. This page publishes its weighting too, and says the same thing about it: averageness 50%, symmetry 30% and proportion 20% are this site's calibration, ordered by which factor the research found strongest, not weights any study has validated.
Why do different attractiveness tests give different scores?
+
Because each one is a different pipeline making different choices nobody discloses. The set of landmark points differs, the weights combining them differ, and the reference set a percentile is drawn from differs, so the same face can land in different percentiles on different sites without either one being wrong about its own math. Underneath all of that, the photo itself moves the numbers: a turned head, a slight smile, harsh side lighting or a wide-angle lens taken close all shift landmark ratios in ways this site's own symmetry and golden ratio pages document with real measured examples. A test that does not state its reference set size or its measurement floor is not more accurate for leaving them out.
What should I wear for my face shape?
+
The usual rule from stylists and opticians is contrast: add length to a round face, width to a long one, and soften a square jaw. For necklines that means V-necks and open necks for round faces, boat and crew necks for oblong faces, and scoop or sweetheart necks for square faces; for shirts, point collars lengthen and spread collars widen. These are conventions, not research findings, and the sources disagree on square and diamond faces. After a scan, this page turns your measured face shape and jaw angle into hair, glasses, neckline and collar directions.
Is my photo uploaded?
+
No. The face model downloads once from this site and runs inside your browser tab, so the measurement happens on pixels already in your device's memory. Nothing derived from your face is written to a server or kept once you close the tab, unless you tick the box for the one free try-on under your result, which sends that photo to Kie.ai to make one image. If you would rather not use your own photo at all, the example button at the top of this page runs the same three factors on an AI-generated example face instead.