Rarest face shape: what 43 measured faces actually show
Short answer: square, once in 43 faces. Longer answer, and the one worth having: we could not find a population study measuring how these six categories are distributed, and the pages quoting percentages do not cite one either. What follows is a count from our own detector, which is a different thing and is labelled as such throughout.
| Shape | Faces | What rules it out, and how often |
|---|---|---|
| rarestSquare | 1 | Ruled out by the jaw angle in 29 of 43 faces |
| Diamond | 2 | Ruled out by the forehead in 32 of 43 — the most concentrated cause of any shape |
| Oblong | 4 | Ruled out by length in 29 of 43 |
| Round | 9 | Ruled out by length in 31 of 43, at the opposite end from oblong |
| Heart | 12 | Ruled out by the jaw ratio in 18 of 43, the weakest dominant cause |
| Oval | 15 | No dominant cause at all: forehead and jaw tie at 16 apiece |
A count of results is not a count of faces
That table says square is rare. Read it once more and notice what it actually measures: how often our classifier returned each label. Between a face and a label sits a set of thresholds, and ours have never been calibrated against faces anyone labelled by hand. The source file says so in a comment next to the numbers.
The clearest evidence that the thresholds, not the faces, are doing the work sits in the last column. Square is refused by the jaw angle 29 times out of 43 — and we have not found a published source that says how many degrees a square jaw is. So square being rare here is most readily explained by our chosen angle being strict, rather than by square faces being unusual. Move that one number and the bottom of the table rearranges.
The honest form of the headline is longer and duller: of 43 faces we measured, with thresholds we chose and have not validated, one was returned as square. Every shorter version of that sentence is claiming more than we can support.
The sex-split version, with its sample size attached
"Rarest face shape for a female" is searched often enough that it deserves a straight answer, and the straight answer is that the sample is small. Twenty-one of our portraits carry a sex in their label. Splitting those gives:
Women n = 12
- Heart
- 8
- Oval
- 2
- Round
- 1
- Oblong
- 1
Square and diamond did not appear once.
Men n = 9
- Oblong
- 3
- Oval
- 2
- Heart
- 2
- Round
- 1
- Square
- 1
Spread across five of the six shapes.
Two things about that before anyone quotes it. Twelve faces cannot tell you anything about women in general — it is a description of twelve portraits. And the label is not a measurement: it records how each portrait was generated, not something read off the image. We have published it because the alternative on this search is a confident percentage with no sample behind it at all, and a small honest number beats a large invented one.
Why there is no real answer to compare ours against
You would expect a question this popular to have been settled by somebody with a large sample. We have not found that study, and the reason may sit further upstream than sample size: the six categories do not appear to have started as an anatomical scheme. They come from eyewear retail, and the number of them has never settled — four, five, six, seven or nine, depending on who is counting. A distribution needs categories that exist before you can measure how often each one occurs.
So every percentage in circulation is a tool counting its own visitors through its own thresholds. Two such counts can disagree without either being wrong, because the disagreement lives in the cutoffs rather than in the faces. We put the side-by-side version of that, along with which confusable pairs have a published measurement behind them and which have none, on the face shape chart.
Physical anthropology does have a real length-to-width scale with published population distributions. It will tell you how long a face is. It will not tell you where oval stops and oblong starts, because nothing maps it onto the styling names — which is the same gap, one level down.
Now measure your own
Reading a description and holding your own face up against it is guesswork, and the descriptions overlap on purpose, because real faces do. The tool measures the four lengths in the diagram plus the jaw angle, in your browser, and shows you how close you sit to every shape rather than picking one and hiding the rest.
Worth knowing before you run it: a rare result here means your measurements sat outside thresholds we chose, not that your face is unusual. The numbers underneath the label are the part that travels.
Front-facing photo, even light, hair off the forehead
Your four measurements will appear here, so you can see which shape they sit closest to.
How the count was produced
Written out because a number without a method is the thing this page is arguing against.
- What was measured. 307 images, of which the detector successfully measured 285. The images are 22 AI-generated test faces, 21 AI-generated portraits used elsewhere on this site, and 264 lighting, crop and compression variants of those 22.
- What the table counts. Only the 43 distinct faces. The 264 variants are excluded on purpose: they are the same 22 faces twelve times over, and including them would multiply a small set into a large-looking one.
- Which detector. The one this site ships, unchanged, run through the same code path a visitor gets. Not a research build.
- When. Re-run on 17 September 2026, after the last change to the classifier, because the previous run predated it.
- What is not here. Any real person. Every image is generated, so nothing in the table describes a human population, and no photograph of anyone was involved in producing it.
Questions people ask about this
What is the rarest face shape?
+
Nobody can answer that from measurement, and any page that gives you a confident number is telling you about its own traffic rather than about human faces. We have not found a peer-reviewed prevalence distribution for these six categories in any population, which is unsurprising given that the categories do not appear to have come from anatomy in the first place. What we can give you is a count: across 43 distinct faces measured with the detector on this site, square came back once and diamond twice. That is a fact about where our thresholds sit, and this page spends most of its length on why those are not the same thing.
What is the rarest face shape for a female?
+
Our portrait subset carries a sex in its labels, so we can split it, but the honest answer starts with the sample size: 12 of the portraits are women. Of those, 8 came back heart, 2 oval, 1 round and 1 oblong. Square and diamond did not appear at all. Twelve faces is far too few to say anything about women in general, and the labels describe how each portrait was generated rather than anything measured from the image, so read that as a curiosity about our set rather than a finding.
Why do face shape sites disagree about which shape is rarest?
+
Because each one is counting its own visitors through its own thresholds, and the thresholds are conventions rather than measurements. Two tools can look at the same face and disagree, because the line between round and oval, or between heart and diamond, is a number somebody chose. Stack that on top of different visitor populations and different category counts — four, five, six, seven or nine shapes depending on the source — and disagreement is the expected outcome rather than a sign that one of them is wrong.
Is oval the most common face shape?
+
It is the most common result, which is a different claim. Oval came back for 15 of our 43 faces, more than any other shape. But oval is defined by not being any of the others: no corner at the jaw, not as wide as it is long, no pointed chin, no narrow forehead. A category defined by absence collects every face without a standout feature, so a tool returning oval most often is the expected behaviour of the taxonomy rather than evidence about how faces are distributed.
Can I trust a face shape percentage I saw somewhere else?
+
Check one thing before you do: does the page say where the number came from? A figure with a stated sample, a stated method and a stated date can be argued with. A percentage with none of those is usually a tool reporting its own output, and the number will move if the tool changes a threshold. That applies to the counts on this page too, which is why the method is written out below rather than left implicit.