You took an attractiveness test, you got a number, and now you want to know what it means. Is 6.2 good? Is 5 average? Why did a different site give you a completely different result?
These are reasonable questions, and the answers are more interesting than a simple benchmark. The short version: "average" depends entirely on how a given tool calibrates its scale, so cross-tool comparison is meaningless — but your percentile within a single tool does carry real information.
Here's how to read your result properly.
Why there's no universal average score
The intuitive assumption is that scores mean the same thing everywhere — that a 6 is a 6. It isn't, for three structural reasons.
1. Different scales, different centres
A rating system's centre point is a design decision, not a discovery. Broadly, tools calibrate in one of three ways:
Normal-distribution calibration. The scale is built so that the population mean sits at the midpoint. On a 1–10 scale, average is 5.5; most people cluster between 4 and 7.
Positively-shifted calibration. The scale is deliberately generous, placing the average around 6.5–7. This is common in consumer apps for obvious commercial reasons — people who receive pleasant scores come back.
Harsh calibration. The scale is deliberately severe, placing the average near 3–4. PSL-derived systems work this way, and present the harshness as rigour.
None of these is more correct than the others. They're different mappings of the same underlying judgments onto different number lines. A 4 on a harsh scale and a 7 on a generous scale can describe identical faces.
2. Different reference populations
Every model is calibrated against some reference set. If the training faces skewed young and the raters skewed toward one demographic, the resulting scale reflects that group's consensus applied to that group's faces.
Cross-cultural agreement about attractiveness is genuinely high — Langlois and colleagues' meta-analytic work in Psychological Bulletin found substantial rater agreement both within and across cultures. But high is not perfect, and residual variation shows up as systematic differences between tools trained on different populations.
3. Different photo handling
Two tools can compute identical geometry and still return different scores, because they handle image conditions differently. Some normalise for head rotation and lighting; some don't. Some warn about camera distance; most don't.
This last one is not a small effect. A 2018 research letter in JAMA Facial Plastic Surgery modelled perspective distortion mathematically: a photograph taken at roughly 12 inches makes the nasal base appear about 30% wider relative to the rest of the face than the same face photographed from five feet. A tool that ignores this is scoring a distorted projection.
The practical rule: only compare scores from the same tool, computed from comparably-taken photographs.
What "average" actually means statistically
Set aside tool-specific scales for a moment and think about the underlying distribution.
Attractiveness ratings, aggregated across many raters, are approximately normally distributed. That has some non-obvious consequences worth internalising.
Most people are near the middle. In a normal distribution, roughly 68% of people fall within one standard deviation of the mean. The bulk of the population occupies a fairly narrow band, and the perceptual differences between people inside that band are genuinely small.
The extremes are rare by definition. Being two standard deviations above the mean puts you in roughly the top 2.3%. The faces most people use as mental reference points — actors, models, influencers — are selected from those tails and then further optimised through lighting, styling, and post-processing. Comparing yourself to that reference set is comparing yourself to a filtered extreme, not to a population.
Small numerical differences aren't perceptually meaningful. The difference between a 5.4 and a 5.8 on most scales corresponds to a difference most observers could not reliably detect. Tools report decimals because the arithmetic produces them, not because the resolution is meaningful.
Percentiles beat raw scores
If your result includes a percentile, that's the number to pay attention to.
A percentile tells you what proportion of the reference population scored below you. It's scale-independent, which makes it the only figure that transfers meaningfully between contexts. "68th percentile" means the same thing regardless of whether the tool uses a 1–10 or a 1–8 scale.
Interpreting percentiles sensibly:
- 40th–60th — statistically typical. This is where most people land, and the perceptual differences across this range are minimal.
- 60th–85th — above average. Noticeable in aggregate, though heavily modulated by photo quality, grooming, and expression.
- Above 85th — well above average on the structural traits the tool measures.
- Below 40th — below average on measured structural traits in this particular photograph. Worth retaking with better photo conditions before concluding anything, since photo quality accounts for a large share of low scores.
Note the repeated qualifier: in this photograph, on the traits measured. That's not hedging. It's the actual scope of the claim.
Why your score changes between photos
People often assume score variation across photos means the tool is unreliable. Usually it means the photos genuinely differ in how faithfully they represent facial geometry.
The main drivers:
Camera distance, as above — the single largest and least-known factor.
Head rotation. Even a few degrees off frontal changes every horizontal proportion the system measures and introduces apparent asymmetry that isn't there.
Lighting direction. Overhead lighting casts shadows under the brow, nose, and chin that alter apparent bone structure. Diffuse frontal light flattens the same features. The face is identical; the measurable image isn't.
Expression. A neutral face and a genuine smile produce different landmark configurations. Most models are calibrated on neutral or mild expressions.
Resolution and focus. Landmark detection degrades on low-resolution or soft images, and degraded landmarks mean degraded measurements.
Here's the useful reframe: the spread between your best and worst photo is itself informative. It quantifies your photographic range — how much your presentation varies with conditions you control. For most people that range is wider than they expect, and closing it is far more actionable than trying to change your structure.
We cover the specifics in how to take the best photo.
What the score genuinely measures
Being precise about scope makes the number more useful, not less.
An attractiveness score estimates how a panel of human raters would rate this specific photograph, based on structural properties with established links to attractiveness ratings: averageness, symmetry, sexual dimorphism, and skin quality. Rhodes's review in the Annual Review of Psychology established through meta-analysis that the first three are attractive in both male and female faces and across cultures; Little, Jones and DeBruine's review adds skin colour and texture.
That's a real signal, and modelling it is legitimate.
What it doesn't capture is the individual layer. Germine and colleagues' twin study in Current Biology found that while some traits are broadly considered attractive, most reliable variation in individual face preference traces to unshared environment — the experiences unique to each person, including the faces they've encountered and their particular social history. The researchers explicitly ruled out measurement error and concluded that individual preferences are genuinely shaped by individual life experience.
So your score is an estimate of the shared component. The individual component — which is substantial, and which determines whether any specific person finds you attractive — is not in the number and cannot be.
How to use your result well
Read the components, not just the composite. A breakdown showing symmetry at the 70th percentile and proportional harmony at the 45th tells you something. A single aggregate number tells you almost nothing beyond rough positioning.
Compare against yourself, not others. Because models are perfectly consistent, they're much better at detecting change than at establishing absolute standing. Same setup, same lighting, months apart, is a fair comparison. Your number versus a friend's — taken on a different phone at a different distance — is noise.
Expect regression toward the middle. Most people score near average, because that's what average means. A middling score isn't a disappointing outcome; it's the modal one.
Don't extrapolate. The research on attractiveness and life outcomes is real — Langlois and colleagues' meta-analyses found that more attractive people are judged and treated more positively — but those are population-level statistical associations with modest effect sizes. They do not predict any individual's experience, and reading a personal forecast into an aggregate correlation is a basic error.
Know when to stop. If you're testing repeatedly, comparing across tools looking for a better number, or feeling worse after each attempt, the tool has stopped providing information and started providing something else. Repeated appearance-checking is a documented pattern in appearance-preoccupation, and it doesn't resolve by testing more.
The honest summary
There is no universal average attractiveness score, because scale calibration is a design choice. Within any single well-built tool, average sits wherever that tool placed it, and your percentile is the figure worth reading.
What the number represents is narrow but real: an estimate of aggregate human ratings of one photograph, based on measurable structural traits. It's genuinely informative about that. It's silent on everything else — including the individual-preference layer that actually determines attraction between specific people.
Curious where you land? Take our attractiveness test — it reports percentiles and component breakdowns rather than a bare number. And for the honest limits of what any such tool can tell you, read how accurate are AI attractiveness tests.
Frequently Asked Questions
What is the average attractiveness score out of 10?
It depends entirely on the tool's calibration. Normally-distributed scales place the average near 5.5; generous consumer scales place it near 6.5–7; harsh PSL-derived scales place it near 3–4. There is no universal answer, which is why percentiles are more useful than raw scores.
Is a 7 out of 10 a good score?
On a normally-calibrated scale, yes — clearly above average. On a generously-calibrated scale, roughly average. Check whether your tool reports a percentile; that figure is comparable across scales and the raw score isn't.
Why do different attractiveness tests give me different scores?
Different scale calibration, different reference populations, and different handling of photo conditions such as camera distance and head rotation. Only compare scores from the same tool using comparably-taken photos.
Why did my score change with a different photo?
Because photos differ in how accurately they represent facial geometry. Camera distance alone measurably alters proportions — a 12-inch selfie makes the nasal base appear roughly 30% wider than the same face at five feet. Lighting, head rotation, and expression add further variation.
Does a low score mean I'm unattractive?
No. It means that in one photograph, on the structural traits the tool measures, the estimate fell below average. Retake with proper photo conditions first. And note that individual attraction depends primarily on the viewer's personal history, which no score captures.
References
- Langlois, J. H., Kalakanis, L., Rubenstein, A. J., et al. (2000). Maxims or Myths of Beauty? A Meta-Analytic and Theoretical Review. Psychological Bulletin, 126(3), 390–423. https://doi.org/10.1037/0033-2909.126.3.390
- Rhodes, G. (2006). The Evolutionary Psychology of Facial Beauty. Annual Review of Psychology, 57, 199–226. https://doi.org/10.1146/annurev.psych.57.102904.190208
- Little, A. C., Jones, B. C., & DeBruine, L. M. (2011). Facial attractiveness: evolutionary based research. Philosophical Transactions of the Royal Society B, 366(1571), 1638–1659. https://doi.org/10.1098/rstb.2010.0404
- Germine, L., Russell, R., Bronstad, P. M., et al. (2015). Individual Aesthetic Preferences for Faces Are Shaped Mostly by Environments, Not Genes. Current Biology, 25(20), 2684–2689. https://doi.org/10.1016/j.cub.2015.08.048
- Ward, B., Ward, M., Fried, O., & Paskhover, B. (2018). Nasal Distortion in Short-Distance Photographs: The Selfie Effect. JAMA Facial Plastic Surgery, 20(4), 333–335. https://doi.org/10.1001/jamafacial.2018.0009