Diamond vs Oval Face Shape: What the Distance Table Says

The Five Checks, Written Out as Widths

Five measurements do the work, and four of the five are widths divided by the width of your cheekbones. In the site's own words they are the face length ratio (face length divided by cheekbone width), the forehead ratio (forehead width divided by cheekbone width), the jaw ratio (jaw width divided by cheekbone width), the jaw angle (the angle at the corner of the jaw, in degrees) and the chin taper (chin width divided by jaw width). Each of its seven outlines gets one centre value on each of those five rows, plus a spread — the amount of variation the site still accepts on that row. A band is that centre plus and minus one spread, and it is what the outline pages print beside each number.

The table below multiplies those published centres out at one cheekbone width so they become numbers a tape can reach: 136 millimetres, this site's working line for face width, named with its band and its provenance on the diamond-face glasses page and not re-derived here. Millimetres and inches in the same cell are the same product rounded twice, not two measurements. Every cell is one multiplication, so if your own cheekbone width is different, replace 136 with your number and the row is yours — choosing one width to print is a presentation decision, not a claim that faces come in sizes.

Read the middle three columns first. At this width the two outlines want a face that is 6.7 millimetres (0.26 in) longer in one reading than the other, a forehead 13.5 mm (0.53 in) wider, and a jaw 11.3 mm (0.44 in) wider. Then read the last two: the variation inside either outline on its own is 24.8 mm on the length row and 16.3 mm on the jaw row — wider than the difference the two outlines are supposed to have on those rows. Only the forehead row has a difference (13.5 mm) larger than its own spread (11.0 mm).

The chin row needs one admission. To turn a chin-taper ratio into a millimetre you have to multiply by a jaw width, and these two outlines do not have the same jaw width — an oval row sits at 109.2 mm here, a diamond row at 97.9 mm. The table prints both rows at a single shared jaw width of 100 millimetres (3.94 in), which is a choice made on this page, not a published number. It is the conservative choice: it holds the jaw still so the difference shown is the chin taper alone. Let each outline keep its own jaw width instead and the difference becomes 19.7 mm (0.77 in) rather than 14.9 mm; hold the jaw at 97.9 mm and it is 14.6 mm (0.57 in); hold it at 109.2 mm and it is 16.3 mm (0.64 in). Four ways of setting the same multiplication, a spread of five millimetres between them, no new measurement in any of them.

In plain terms: of the five rows you can put a tape against, one of them puts a bigger gap between these two outlines than the everyday variation inside a single outline, and it is the forehead. The angle row is not a width at all, so it is printed in degrees, and its difference is just over two degrees against a spread of seven and a half.

Ratios used, printed the way the rule-set page prints them: length 1.45 against 1.40, forehead 0.90 against 0.80, jaw 0.80 against 0.72, angle 122° against 120°, chin 0.45 against 0.30. The millimetre cells come from the unrounded centres (for the forehead row, 0.89926 and 0.80), so multiplying by the rounded 0.90 gives 122.4 mm rather than 122.3 mm — that tenth of a millimetre is rounding, not measurement. Spread cells are one published standard deviation of that row at that width: 0.18228, 0.08088, 0.12000, 7.65957° and 0.08333 of a ratio unrounded, printed on the rule-set page as 0.18, 0.08, 0.12, 8° and 0.08. The shared-width column is the overlap of the two ±1σ bands, worked from unrounded band edges; at the forehead row the printed edges are 0.82 to 0.98 and 0.72 to 0.88, which overlap by 0.0625 of a ratio, and 0.0625 × 136 mm is that 8.5 mm. One spread is a chosen band width, not a natural boundary: widen it to two and the ground both rows accept at this cheekbone width becomes 92.4 mm on the length row, 30.5 mm on the forehead row, 54.0 mm on the jaw row, 18.4 mm on the chin row and 28.60° on the angle row — which is why the table stays on one band and says so. The chin row's 100 mm jaw is set by this page; at each outline's own published jaw width the same centres give 49.1 mm (1.93 in) for oval and 29.4 mm (1.16 in) for diamond, a difference of 19.7 mm.

Only Two of the Five Rows Have Real Daylight in Them

Differences in millimetres cannot be compared across rows, because a millimetre of face length and a millimetre of chin are not the same achievement. There is one way to put all five on the same footing, and this site already uses it for its distances: divide each difference by that row's own spread. The row's spread is how much that measurement moves between one face and the next inside a single outline, so the result says how many units of ordinary variation separate the two published rows. It has no unit, which is the point — the same arithmetic works on a ratio and on a degree.

Do that to the five rows above and you get 0.27 for length, 1.23 for the forehead, 0.69 for the jaw, 0.27 for the angle and 1.79 for the chin. Set against the mark of 1.00, one full spread, the count is two rows out of five: the forehead row and the chin row. The other three leave the two outlines closer together than the variation inside either of them, which is why their bands overlap so widely in the last column of the first table: 42.9 mm of shared face length, 21.3 mm of shared jaw width, 13.28° of shared jaw angle.

And here is the part that decides what you can do with this at home. Of those two rows that clear the mark, the chin row is the one a soft tape cannot give you. Where the narrowest point of a chin is, is ambiguous by more millimetres than the whole shared strip is wide, which is why this site's calculator has no field for it and why that route runs with the row's contribution switched off rather than with a guess put in its place. That limitation is stated on the diamond-versus-heart page and on the inverted-triangle guide; nothing is added to it here.

So the honest count for anyone working from their own measurements is one, not two: exactly one check puts more than a spread of daylight between an oval row and a diamond row, and it is the one you can take with a tape — forehead width divided by cheekbone width. If you have a photograph read instead, all five rows are supplied and the chin row becomes the widest gap of the five.

The Distance Table: Twenty-One Pairs, One Number Each

This site classifies against seven outlines, and any two of those seven make a pair, so there are twenty-one pairs to compare. Each pair gets one distance, built from all five rows at once: every row-gap is put in units of that row's own spread, each row is weighted by how much this site's calibration says it is worth, and the five weighted gaps are combined into a single number. It is the same measure the bars beside your result are drawn from, computed here for every pair rather than only for yours.

The table is sorted from closest to farthest, and it carries one extra column the site does not print anywhere else: for each pair, which of the five checks contributes the most to that distance, as a fraction of it. That column is why this page can say something more useful than the ranking — it is what lets you see that the pair at the top of the table is not close because it is identical, but close because one specific check does nearly all the separating while four do very little.

Read the top two lines together before anything else. Oval and diamond: 1.9325. Oval and oblong: 1.9539. The difference between first and second place is 0.0214, which is about one fortieth of the distance itself. Whether you are willing to call a race won on that margin is the question section four answers, and the answer is not a simple yes.

One sentence to take away: the shortest distance in the table is between these two outlines, the longest is heart against triangle at 4.6993, two and a half times as far, and the margin by which the first place wins is smaller than most of the numbers in the table.

Distances are this page's own calculation from the published constants, run against the same function the result pages call, and printed to four decimal places; the unrounded value for the top pair is 1.932456791960106. The list is a table of the rule set, not a table of faces: it says how far apart seven published rows are, not how distinguishable two people are. Nothing in it is a prevalence figure, and this site publishes none, for the reason printed on every outline page — the figures in circulation state neither sample nor method. Three rows deserve a caution when you read the carries-most column: heart and diamond sit 0.99 of their distance on one check, oblong and square sit 0.91 on one check, and both pairs still land well down the table, which is what a single decisive row looks like when it is not enough on its own.

What Closest Depends On, and Where It Stops Being True

A distance is a weighted one, and this site publishes its weights, so the honest thing to do is re-run the whole table with other weights and report what moves. Four settings are printed below. Two of them are weight sets this site actually runs: the published one, and the one the tape-measure route uses, which drops the chin row because a tape cannot supply it and estimates the jaw angle from the three widths instead of measuring it. The third adds the boost that route gives to a jaw angle you measure yourself. The fourth is not published here and is printed only to test the claim: it gives all five checks equal weight.

Under the published weights the pair leads by 0.0214. Under the tape route it leads by 0.4399, twenty times wider, and under the self-measured-angle variant by 0.4585. That widening is not a rounding effect and it has one cause: the row the tape route drops is the row carrying 0.42 of this pair's distance, so removing it moves these two outlines far more than it moves the runner-up, whose distance rests on the length row. The route a reader can actually take at home therefore makes this the clearest winner in the table rather than the closest call.

Under equal weights, though, the claim does not survive. The pair's distance rises to 2.3091 and it drops to second place, behind oval and oblong at 1.7946. So state it precisely: among the numbers this site publishes, and under both weight sets it publishes, oval and diamond are the closest pair of its seven rows — and the margin that makes it first place is the weighting, not the raw centres.

What you can argue with: the ranking is a property of how this site weighs the five checks, and this site says what those weights are for, so you can re-run the question with weights you prefer. The first table is the place to start, because its gaps are the same five you just read.

The angle boost in the third line is a fixed multiplier of 1.5 on the jaw-angle weight, applied when you measure the corner yourself rather than have it estimated. The site states plainly that this multiplier has no paired data behind it — the calibration set contains no case of one person both measuring and estimating their own angle — so it is a declared assumption rather than a fitted value, and it is printed here as one. The margins are differences between two distances in the same run, so they are in the same unit as the distances. The equal-weight line is the only one of the four whose ranking changes; nothing else in this table moves rank.

Which Check Carries the Distance

The distance between these two rows is not a tie spread across five checks. Split the number that produced it and each row's share falls out, and the split is where this page earns its keep, because it settles an argument the site has been having with itself in three different places.

Read the shares from largest down: the chin taper carries 0.42 of it, the forehead ratio 0.39, the jaw ratio 0.14, the face length ratio 0.03 and the jaw angle 0.02. Two checks carry eight-tenths of the distance between these two outlines; three carry almost nothing.

Now the three published orderings. The pair is registered in this site's confusion table with the forehead ratio and the jaw ratio named as its features — those are the two rows the written explanation for this border is about. The outline page, in its own table of what gets confused with what, points at the chin taper as the sharper test, and the paragraph above it says the chin taper comes first and the forehead second, and adds that the jaw angle is the feature doing the least work. And this page's share column puts them in the order chin, forehead, jaw, length, angle.

So the outline page's ordering and the computed shares agree, position by position, and the registered pair is a different kind of statement: it records which border has copy written for it, and it does not claim to rank the rows by how far apart they sit. Nothing here needs the site to pick one of those and disown the other. It needs a reader to know which kind of sentence they are holding — and that is exactly the sentence the result page beside your photo cannot tell you, because it prints a named boundary, not a share.

In one line: two checks do almost all the separating on this border, the widest of the two is the one a tape cannot give you, and the narrow one (the angle at the corner of your jaw) is the one worth not re-measuring three times.

Centres are printed on the tape-measure scale, the one the rule-set page publishes and the one a soft tape measures against, to the site's usual two decimals. The gap column divides each unrounded difference by that row's unrounded spread, which is why it can be compared across rows given in ratios and in degrees, and why the same gap number appears whichever of the site's two scales you start from. Weights are the published five, normalised so their average is 1. Shares are each row's weighted gap as a fraction of the five added together: 0.1166, 1.4610, 0.5272, 0.0604 and 1.5691 over 3.7344, whose square root is the distance.

Why the Row That Carries Most Is the One Weighed Least

The chin taper has the widest gap of the five and the smallest weight of the five. That looks like a mistake, so it is worth explaining the mechanism rather than defending it.

The weights are not preferences. They came out of the calibration set as a ratio, for each check, of how much that measurement varies between the labelled groups to how much it varies within a group. A ratio above 1 means the row tells the groups apart; below 1 means it varies as much inside one outline as between two, so it is not carrying information about which outline a face belongs to. On that reading the five rows came out in the order length, jaw, forehead, angle, chin — with chin taper's ratio below 1, at 0.84, which is why it ended up at 0.49, the smallest of the five.

Then the admission the site makes about that number in its own notes: it is set to 0.49 and not to 0, because on this site a weight of 0 does not mean useless. It means this path cannot measure the row at all, which is what the tape route does with the same row. A low weight and a zero are two different statements, and the chin taper is a low weight on the photograph path and a zero on the tape path.

That is the whole reason the shares come out as they do: the row with the widest raw gap gets the least credit, so the distance between these two outlines stays small even though the two ends of the face are genuinely different. It is also why the tape route, which deletes that row, makes this pair the clearest winner in the table instead of the closest call.

The awkward part: the most visible difference between these two faces is the one this site's own data says is least useful for telling faces apart, and the site prints that as a weight you can argue with rather than hiding it in a total.

What This Site Already Prints About This Border

Before adding to the record, it is worth saying how much of the description is already published here, because the answer is: the words, in several places, and the arithmetic in none of them until this page.

The outline page for this shape carries a card for this pair and points at the chin as the deciding row, with the sentence that an oval chin curves while this one comes to a point, and a second paragraph naming the order it thinks the checks decide in. The oval page carries the matching card from the other side, opening with the line that both outlines are longer than they are wide and both narrow toward the chin, and closing with the instruction that if your chin taper reads high the outline is oval rather than diamond. Neither card carries a number.

That page also answers the question the quiz asks — how do I know if my face is diamond or oval — in prose. The glasses page for this shape carries a table whose note names oval as the rule set's boundary for this shape and heart as the pair the forehead alone decides, and further down answers why size charts give a diamond face the same numbers as an oval one. The two outline pages link to each other as nearest boundaries, and both result pages say the same thing about this pair — see the next section.

So the pair has been described, taught, sized and linked from five pages. What has not been printed anywhere on the site, until the table above, is the distance for every pair so you can see where this one ranks; the share each check contributes; or what happens to the ranking when the weights change. Those three are this page's contribution, and the millimetre table at the top is a multiplication of rows the site already publishes.

Two Result Pages That Name Each Other

This site does not hard-code which boundary is the tightest. It computes it, from the same published constants, every time a page is built. Run that over the seven outlines and the answer comes back oval and diamond, at 1.932456791960106, with the runner-up oval and oblong at 1.9539. So when a diamond result names oval as its nearest boundary, and an oval result names diamond, those are not two editorial decisions that happen to agree. They are one calculation reached from both ends.

Both pages print the same sentence with their own name first, and it reads: the nearest boundary in our rule set is Oval and Diamond, the two prototypes differ mainly in the forehead ratio and the jaw ratio, and if your result feels wrong that is the pair worth reading about before re-measuring. Note which two checks that sentence names — forehead and jaw. They are the pair registered for this border, and they are not the two rows that carry the most distance, which on this pair are chin and forehead. The boundary text is about what the site has written an explanation for. The share table above is about what moves the number.

There is a reading trap in that printed sentence, and since this page is about reading it, it belongs here. The two numbers in it are printed in the order the registered pair stores them — this shape's value first — while the two names at the head of the sentence are printed in the order the page reads them. On a diamond result those two orders coincide. On an oval result they do not: the sentence opens Oval, then Diamond, and the numbers that follow are the diamond value then the oval value. Read the pair of numbers against the pair, not against the order the names arrived in.

If you came here looking for oval vs diamond face shape because a result disagreed with what you see, the runtime has a specific behaviour worth knowing. Where two of this site's own paths return different outlines for this pair, the page prints a note naming the border, quoting the two rows it says differ, and adding that the pair is the closest in the rule set — with links to both outline pages. Where a pair is not registered, the same mechanism refuses to invent an explanation and tells you to re-take whichever measurement was quicker instead. This pair is the registered one, so it does get the note.

And one limit on what you will see on those pages: the result page shows your own five readings, the row it scored them against, and the distance bars. The twenty-one-pair table and the share column above are not printed there and are not going to be moved there. They exist on this page so that a reader who wants to know how much room this border has can look it up once, in one place, rather than being handed a boundary name with no scale.

What the Pages That Answer This Question Print

Searching this pair brings up a page whose title is exactly the question, and it answers with a four-row table of descriptions. Face length for oval reads longer than it is wide, and for diamond narrow forehead and a small chin. Cheekbones: oval the widest part of your face, diamond significantly wider than your forehead and jawline. Jawline: oval no sharp angles and more of a rounded jawline, diamond a narrow chin that tapers to a strong point. And the forehead row gives both outlines the identical sentence — the forehead's width is smaller than the width of your cheekbones. That page also states three times over that these two are the shapes people confuse, which is the opposite of a claim nobody has made.

The pair gets noticed, and the same page that notices it prints one of the five rows twice. It is not carelessness. It is what happens when a border is drawn with words: the forehead row is true of both outlines, so it cannot be the difference, and the description has to carry the weight in adjectives instead. A second page does the same thing with inequalities — oval given length greater than cheekbones and forehead greater than jaw, diamond given cheekbones greater than forehead greater than jaw, and the widest point named as cheekbones for both — followed by a set of inch measurements per outline with no difference printed between the two rows.

Elsewhere the numbers that appear are on other borders. One reader's own discussion entry gives a length-to-width ratio for oval and a higher one for oblong, and says of diamond only that it is the rarest and most photogenic, with a narrow forehead and jaw and prominent cheekbones. A clinic's guide gives oval's length as about one and a half times its width, in words rather than as a number, and names the outline that gets mistaken for oval as the long one. A jeweller's page in the top ten is about diamond cuts; the word collision is the reason it ranks, and its six percentage signs turned out to be discount copy.

Three pages could not be read at all on the day and are quoted here from the summaries only, with that limit marked in the table below. Two of the three place this pair on other borders as well: one beauty guide's own main comparison is diamond against heart rather than diamond against oval, and one magazine's sentence on the pair is that diamond faces are similar to oval ones in that the cheekbones and centre of the face are widest, even if marginally. That is the closest any reachable page comes to naming where these two actually differ, and it carries no number either.

Above all of that sits a block generated for this question rather than written by anyone, and it is worth naming what it does to the arithmetic, because it is the most-read answer of the set. It offers the reader a choice of body parts rather than a comparison — one width or the other being wider, as alternatives — and gives a single number in its entire comparison, a length of about one and a half times the width. Which width is wider is the check itself, so writing them as alternatives removes the only part of the sentence you could test. That block is a rewrite of the pages under it, and it is not counted as a position in the table below.

Every Dated Position on This Border, Oldest First

The rows below are every position found for this question that carries a date, oldest first, including this site's own. Identities are given by type rather than name for every row that is not this site. Nothing outside this site's rows puts a number on the distance between these two outlines, and nothing outside this site's rows publishes a band.

Identities are given by type, not by name or link, for every row that is not this site, and no row here is offered as a comparison of this site against anybody else. Dates are the publication date printed on the page, with the revision date where the page carries one; two of the external rows have no readable body and their dates come from the result listing, which is stated in those rows rather than quietly used. Three rows share the publication date 2026-09-12 — this site's diamond guide, oval guide and result pages — and no order inside that date is claimed; they are listed by the date each was last revised. Not in this table, and not evidence for any sentence above: one reader-written discussion page and one question page that carry only relative timestamps, one jeweller's page whose printed date turned out to belong to a sidebar card rather than to the article, one idea-collection page with 151 words and no article behind it, and one short video. The generated answer block at the head of the results is a rewrite of the pages above it and is not counted as a source.

What Moved Between 2018 and the Newest Page Here

Six dated external positions, eight years and three months between the first and the last, and the format has not moved: not one of them prints a distance, a band or a tolerance for this border. What did move is that somebody finally put the pair in a title — the newest external page in the set is the first reachable one whose main heading is this comparison — and having made that move, it answers with four rows of descriptions and gives one of the five rows the same sentence twice.

So the gap between those pages and this one is not that they missed the pair. They named it, and two of them said plainly it is the most-confused pair on their list. The gap is that nobody has given the border a scale. Adjectives, inequalities and per-outline example widths are all statements about ordering, and ordering is the part everybody agrees on: this site's rows agree with theirs that both outlines are narrower at the forehead than at the cheekbones, and both are longer than they are wide. Where the agreement ends is at the question of how much, and after that at the question of how much is worth anything.

That second question is the one this page can answer and the others cannot, because it needs a spread as well as a centre. A 6.7 millimetre difference in face length sounds like a difference until you see that the variation inside one outline on that row is 24.8 millimetres; a 2.04 degree difference in jaw angle sounds like nothing until you notice it is a quarter of a spread, which is exactly as much as the length row's difference is worth, and both are dwarfed by the chin row. Nothing about the ordering changes if you fix the scale. What changes is which of the five sentences you would stake a decision on.

The Order to Run It In

Take four measurements yourself and this site's calculator scores them against all seven rows on the scale section one is printed in — forehead against cheekbones included, chin row omitted, because a tape cannot supply it. Face shape calculator ↓

If you want a usable answer rather than another description, run this pair in a fixed order, and the first step is the one the reachable pages skip because they have nothing to check it against. One: establish what each number you are holding is divided by. A tape gives you widths; a photograph reader gives you its own values, which are not the same quantities even when the label matches. Two: compare the forehead against the cheekbones, since that is the only row on this border with more than a spread between the two outlines that you can supply yourself. Three: only if that row comes out level, look at the jaw row and the length row, knowing both of them sit inside ordinary variation between two faces of the same outline. Four: leave the chin row to a photograph reading, because at home there is nothing to multiply by — the jaw width you would need beside it is the number your tape is least reliable at.

For step two, the useful question to ask any page, this one included, is not what its number is but what its number is worth. This site's rule-set page prints the five centres and the five spreads, so the division is one you can do in your head with the page open. If your forehead divided by your cheekbones sits near 0.90 you are on the oval side of this border; near 0.80 you are on the diamond side; between 0.82 and 0.88 both rows accept you, which at the working cheekbone width is a strip 8.5 millimetres across.

For a diamond vs oval face decision you intend to act on — a haircut, a frame, a reply to someone about their own face — the two rows that separate these outlines are worth more to you than the outline name. Section six is where the reason sits: the row with the widest visible difference is the one this site's data rates least useful, so a label that leans on it is leaning on the softest of the five.

What This Page Does Not Prove

Nothing here establishes that an oval outline or a diamond outline is more attractive, more common or morephotogenic. This site publishes no prevalence figure for any of its seven shapes, for the reason it prints on every outline page: the figures in circulation state neither sample nor method. A rarest-face-shape claim appeared in one of the pages read for section nine, and it is not answered here, above it or below it.

It does not prove that these two outlines are hard to tell apart in general. It says that two published rows of one published table sit 1.93 weighted spreads apart while four others sit between 1.95 and 4.70, and that the number rests on a weighting this site chose and prints. Equalise the five weights and the pair drops to second in its own table; that result is in section four rather than hidden in a footnote because it is the most useful thing on this page after the table itself.

It is also a distance between two rows of a rule set rather than a measurement of two faces, and the difference matters. One of these two rows came out of eighteen measured photographs. The other did not come from a measured group at all: three photographs in the reference set carried the diamond label, and the centre of that row was placed by interpolating along the ordering the older literature gives these outlines, which this site says on its own result pages and repeats here. Every number on this page that touches the diamond row inherits that, including the 0.42 share and the 1.9325.

And the twenty-one distances are this page's calculation, not a finding about people. They are reproducible from the published constants and they are printed to four decimals so you can check them, but they describe how this site's table is shaped. If you disagree with the weights, the first and third tables are the two places to argue — they carry the gaps and the weights separately for exactly that reason.

Widening the band from one spread to two, as the first table's footnote shows, takes the forehead row's shared strip from 8.5 to 30.5 millimetres and the chin row's from 1.8 to 18.4, which is enough to swallow most of what separates these two rows in practical terms. Any arithmetic here can be redone with your own widths and your own two readings in place of the published ones. If it stops being useful when you do, your face is the reason, and the arithmetic is not arguing with it.

Questions this page answers