The pretty scale, and what its numbers really mean

Nowhere in particular, which is the first thing worth knowing. There is no research programme behind the 1-10 face rating and no standards body that defines it. It spread through school corridors, magazine culture and then internet forums, and it stuck because it is easy to say and impossible to check.

When psychologists study attractiveness they do use rating scales, but they use them very differently. Ratings are collected from many people, averaged, and reported only as group means - because an individual rating is known to be unreliable. The published result is never "this face is a 7". It is "this face averaged 4.8 across 62 raters, with a standard deviation of 1.6", and that standard deviation is the interesting part: it is large, consistently, in every study.

Studies that separate shared taste from individual taste find the split runs close to even, and in several large samples the individual component is the larger one. That means when someone rates a face, up to half of what they are expressing is specific to them - their history, their preferences, who the face reminds them of - and cannot be predicted from the face at all.

This is why a single number on a 1-10 scale carries so much less information than it appears to. The number is a summary of one person's reaction, and a different person would produce a different number from the same face without either of them being wrong.

A pretty-scale calculator cannot access anyone's reaction, so it measures geometry and maps the result onto the familiar scale. The measurements are usually some combination of:

• Bilateral symmetry - how closely the two halves of the face mirror each other.
• Facial thirds - whether hairline-to-brow, brow-to-nose and nose-to-chin divide evenly.
• Facial fifths - whether face width divides into five equal eye-widths.
• The golden ratio - how close face height to width sits to 1.618.

Those are real measurements of real things. The step that is not justified is the last one: converting them into a rating and presenting it in the same format people use socially, which invites you to read a geometry score as a social one.

The score here is out of 100 and lands in one of five bands. The breakpoints are read off a committed reference distribution of 7,997 measured faces, and each one is the lowest score whose whole percentile plateau clears the claim it carries, so the population claims are literally true for everyone in the band. The bottom two bands carry no population claim at all - a rank is useful information at the top and only discouraging at the bottom.

What is the pretty scale?

In everyday use it is the informal 1-to-10 rating people apply to faces. It has no standard definition, no reference points and no agreed midpoint, which is why two people using it on the same face routinely differ by two or three points. In software, a pretty scale is usually a facial-geometry calculator: it detects landmarks, measures symmetry and proportions, and maps the result onto a 1-10 display so it reads like the familiar social scale.

Is the pretty scale accurate?

The measurement underneath a software pretty scale can be accurate - landmark geometry is precise and repeatable. The scale itself is not, because there is nothing to be accurate against. There is no calibrated population standard for what a 6 or an 8 means, so each tool picks its own mapping. That is why the same photo scores 5.8 on one site and 7.4 on another without either being wrong on its own terms.

What is the average score on a pretty scale?

It depends entirely on the tool, which is the problem. Tools that map to a normal-looking curve usually put the median around 6 to 6.5 out of 10, because a median of 5 makes half of all users feel below average and stop using the product. The distribution is a design decision, not a discovery.

Why does this site score out of 100 instead of 1-10?

Because a mark out of 10 is read as a rating of a person, and the tool is measuring a photograph. The 1-10 format also carries the social scale's baggage - everyone already believes they know what a 4 means. A score out of 100, labelled facial harmony, is harder to mistake for a verdict and closer to what the arithmetic actually produced.

What is a good pretty scale score?

On this site the reference distribution is 7,997 measured faces, and it puts the median at 74 out of 100, with half of all faces between 65 and 80. A score of 83 or above is the top 15% and 86 or above the top 5%. On other tools you would need their published distribution to answer this, and almost none publish one.

Can I convert my score to a 1-10 rating?

Arithmetically, yes - divide by ten. Meaningfully, no. The 1-10 social scale is not linear and not shared: people cluster their ratings around 5-8 and use the ends rarely, so a conversion implies a precision the underlying judgement has never had.

• How pretty am I? The main test
• Beauty scores and how they are calculated
• The full method behind this score