Round Glasses for a Square Face: the Width in Millimetres Decides It

The millimetre table, and the one number that is not on the frame

The five numbers that decide whether a frame fits are not the five numbers printed on it, and the gap between the two sets is where most square-face round-frame purchases go wrong. Three are stamped on the inside of the temple arm — lens width, bridge width, temple length, in that order, written either with dashes or with a small box symbol between the first two, so that 52-18-145 and 52□18-145 are the same frame. Two more you have to bring yourself: your pupillary distance, which comes from the refraction and never from the frame, and the frame's total width, which nobody prints because it is the sum of the stamped three plus the two end-pieces, and the only way to get it is to lay a ruler across a frame that already fits you. The table gives all five, what each comes to on an adult face, and what each has to match.

Which numbers here are published and which are ours should be stated. The three temple-arm ranges are the standard box-system reading, and two frame-sizing guides agree on the lens width — 40 to 62 mm — while differing on the other two: one prints bridge 14 to 24 and temple 110 to 150 mm, the other bridge 14 to 28 and temple 120 to 150. The narrower bands in the table above are our own narrowing of those to what an adult face actually uses, and are not published figures. The total-width bands against face size — small under 129 mm, medium 130 to 139, wide over 139 — are printed by an English frame retailer, which also states the rule the first row runs on: the total width should be similar to the width of the face, and the lower edge should not sit on the cheeks. The 5 to 8 mm the second section adds for the two outer rims is published by an English frame-size chart, whose formula is twice the lens width plus the bridge plus that allowance. The adult PD figures are published twice and disagree in the way ranges do: one eyewear retailer gives 54 to 68 mm as the average adult range, and an eye-care editorial source gives 60 to 65 mm for adults aged 19 and over, which is the narrower figure this page works to. The decentration ceiling is the one an English optical-education source states: keep it under 5 mm per eye, with 3 mm the comfortable target. Ranges are adult averages, not limits; a face outside them needs the formula run on its own PD rather than the band read straight across.

The one formula that is not a matter of taste

Frame PD is the lens width plus the bridge width — the distance the frame puts between the two optical centres. Decentration is how far each lens then has to be shifted inward to land on your pupil: (frame PD − your PD) ÷ 2, in millimetres, per eye. A frame whose two centres already sit where your pupils do needs no shift and comes out at zero. Every millimetre of shift is a millimetre of extra edge thickness on a minus lens, which is why the number has a ceiling. Under 3 mm is comfortable. Three to 5 mm is wearable, and is where most off-the-shelf frames land. Over 5 mm the lens is decentered far enough that both the thick outer edge and the soft, blurry periphery of the lens survive into your field of view, and the frame is worth putting back no matter how it looks on your face.

Run it once and the PD-to-lens-width bands stop being a table somebody published and become arithmetic you can do for your own number. Hold the bridge at a typical 18 mm and aim for a 3 mm decentration, and lens width comes out at PD − 12; let the decentration reach its 5 mm ceiling and lens width reaches PD − 8. So a 60 mm PD wants a lens between 48 and 52, a 65 mm PD between 53 and 57, a 56 mm PD between 44 and 48. The reason a low-PD wearer keeps being handed frames that are too wide is that the stamped lens number was chosen for a 64 mm face, not theirs — and on a square face the trap has a second jaw, because the face already reads widest at the cheekbone, so an oversized round frame is the one shape that can make the width worse while it is trying to soften the angle.

One published set of bands in the table below runs wider than this, and the gap is worth working rather than waving away, because it looks like a contradiction and is not. The fitting guide dated August 2026 hands a 60 to 65 mm PD a lens of 50 to 53 mm; the arithmetic above hands the same PD 48 to 53 at the 3 mm target. Take its lower figure and run it: a 50 mm lens on an 18 mm bridge at a 60 mm PD decentres by (50 + 18 − 60) ÷ 2 = 4 mm, which is past the 3 mm ideal that guide itself states and inside the 5 mm ceiling it also states. The two are not disagreeing about anything; they are aiming at different points of the same range, and the difference is one millimetre of edge thickness per eye. Where a published band and this page's derivation part company, the derivation is the one to follow, because it tells you which point of the range you have landed on instead of only that you are inside it. The same row also tops out at a 65 to 70 mm PD, above the adult range the millimetre table lists, which is the ordinary reason a chart and a measurement disagree: the chart is covering faces the measurement leaves out.

Now the number that is not on the frame, and is the one this keyword turns on. The box numbers give you the frame PD — lens plus bridge — which is not the total width. Total width is twice the lens width, plus the bridge, plus the two outer rims beyond the lenses, and an English frame-size chart puts those rims at another 5 to 8 mm. So a 52-18 frame, whose frame PD is 70, measures 52 + 52 + 18 = 122 mm across the lenses and bridge and somewhere around 127 to 130 mm in total. Hold that against the bands in the table above and the useful fact appears: an average stamped frame lands right at the boundary between the small band and the medium one, under 129 mm or just over it, while a square face's cheekbone width usually wants the medium band or the wide one above it. That is why the round frames bought for this shape so often come up short — not because round is the wrong family, but because the size chosen was the size the stamp suggested rather than the size the face measured. Get the total width onto the cheekbone first and the round-versus-oval argument mostly settles itself, which is the part the shape advice never tells you.

A decade of square-face advice, by the date it was published

The shape verdict for a square face has moved three times in ten years; the fitting arithmetic has not moved at all. Laying the published advice out by date is the only way to see the two separately, because in any single article they are blended — a stylist's “round softens the jaw” sitting beside an optician's decentration ceiling as though the two were the same kind of claim. They are not. One is an opinion about line, and it swings. The other is arithmetic about where a pupil sits in a lens, and it has been identical in every year below that bothers to state it. Names are withheld on this table by the same convention the rest of this site uses for its comparison tables; the dates are publication dates, and every number in the third column can be checked against the formula above without needing the source at all.

Fourteen positions across ten years, one row per published position, dated by when it appeared. Read the fourth column down and the shape verdict moves three times: “avoid a true round” through 2019, “oval or round-within-square” through the early 2020s, “round, oval, cat-eye or half-frame, just not a hard square” by 2023 to 2025 — and then, in 2024 and again in the 2026 fitting guide, a pull-back to “it depends which square face you have,” with the certified guide landing on oval or inverted-trapezoid rather than round. Read the third column down and nothing moves: every row that carries a number carries the same decentration formula and the same 3-to-5 mm ceiling, in 2021, in 2025 and in 2026, stated identically. The shape advice is an opinion and behaves like one; the fitting math is arithmetic and behaves like that. Names are withheld by this site's convention for comparison tables and the dates are publication dates, so the load-bearing claim is the divergence between the two columns, not any single row — and the third column can be re-derived from the formula in the section above without any of the sources.

What did not move, and why that is the half to trust

The divergence is the useful part, and it tells you which half of any article to believe. When a stylist's column and an optician's worksheet disagree about whether a square face should wear a round frame, the disagreement is real and unresolvable, because it is about line and line is taste — ten years of it above, three verdicts, no referee. When they agree about decentration, they agree because there is nothing to disagree about: (frame PD − your PD) ÷ 2 is either under five or it is not. So the shape question — round, oval, cat-eye, soft-square — is the part you are allowed to have an opinion on and change your mind about, and the width question is the part you are not.

Most of the advice stacked against this keyword argues the first question and quietly leaves the second one out, which is why it can tell you that round softens a square jaw and still hand you a frame 7 mm decentered and too wide for your PD to wear. That is not a small omission. The shape claim is the one that gets repeated because it is easy to say and impossible to check; the width claim is the one that decides whether the glasses are comfortable, and it is buried at the bottom of a fitting guide if it appears at all. Read the third column of the table again: it is mostly em-dashes. A decade of people telling a square face what shape to buy, and the numbers that would tell it what size to buy appear in four rows out of fourteen.

When round goes too round

This site classifies a face on five readings, and a square face is square because of two of them. The jaw angle — the angle at the corner below the ear — is centred at 127.9° on this outline, the lowest of the seven the rule set carries, which is the numerical form of “the line turns instead of curving.” The jaw ratio, jaw width against cheekbone width, is centred at 0.81, higher than on any outline except the triangle, which is the numerical form of “the jaw is nearly as wide as the face above it.” Those two are the decisive readings, and a frame is worn across both: its lower rim runs along the jaw angle, and its total width lands on the width line the jaw ratio describes. The other three do not separate this shape from its nearest neighbour. Face length ratio is centred at 1.103 and chin taper at 0.44 — the same two values the round face carries, digit for digit — and the forehead ratio sits at 0.95 against the round face's 0.93. That is why a square and a round face read as cousins until you find the corner, and it is worth knowing that the corner is the harder of the two to measure: on the calibration photographs behind this site's rule set the jaw angle did not separate square from round cleanly, which is one reason the result page reports your margin against all seven outlines and a confidence, rather than a single name.

A round frame works against the jaw angle and only the jaw angle. It puts a continuous curve above the cheekbone, so the eye reads a curved line where the bone underneath turns a corner, and the corner reads softer. It does not move the angle — the angle is bone, and a frame moves no reading, it only changes which way the eye reads it — but a curve set against a corner is the whole mechanism, and it is the same mechanism the rest of this site's square-face advice runs, from hair that falls in a wave beside the jaw to a beard line kept rounded rather than traced along the corner. What a round frame does not touch is the width. The width is set by total frame width against the cheekbone, from the table above, and a round lens with the right curve but the wrong width still leaves the jaw as the widest line. This is why “round softens square” is half an instruction: it names the curve and stays silent on the width, and the width is the half that can fail.

Round goes too far in three directions, and each has a number on it. The first is too narrow, and it is the commonest. A round lens is usually cut small, because a true circle reads as a statement at any size, and a small circle on a face whose cheekbones are wide lands inside the width line — total frame width under the cheekbone measurement — and leaves the jaw as the widest thing on the face. That does not soften the square reading; it strengthens it, by removing the only line that was competing with the jaw. The check is the ruler from the table: total width at the cheekbone, never inside it.

The second is too deep. A circle is as tall as it is wide, so a 50 mm round lens is also 50 mm deep, and depth is the dimension a square face can least afford — past roughly half the face length the lower rim drags the mid-face down, and on a high myope the bottom of a tall lens is its thickest part, worst with astigmatism near the 180 axis. The 2025 dissent in the table makes the same point from the other side: a large frame does not make a square or short face smaller, it makes it read shorter and more square. The check is the lower-lid-to-nostril distance divided in thirds; the frame's lower edge belongs in the middle third, not the bottom one.

The third is the wrong sub-shape, and it is the one the keyword hides. A face that reads round from the front but still turns a corner at the jaw from the side — high on the width match, low on the jaw angle only in profile — does not want a true circle. A true circle echoes the front-roundness and erases the side-corner, and the result is a face that looks smaller and slightly costumed. That face wants round-within-square: an oval, or a soft rectangle with rounded corners, which carries the curve the jaw angle needs without giving up the line the front view already has. This is the same front-versus-side test that decides square from round in the classifier, read on a frame instead of a bone.

So the answer to the keyword is conditional, and the condition is a number rather than a slogan. A square face is not buying “round.” It is buying a lens with more curve in it than the jaw underneath, held to a total width that reaches the cheekbone and a decentration that stays under five and ideally under three. Inside those bounds round, oval, cat-eye and soft-square all do the job, because all of them carry the curve the jaw angle is short of and none of them has to be a true circle to carry it. Outside them — a circle too small to reach the cheekbone, too deep to clear the mid-face, or too round for a face that is only round from the front — the same word “round” describes a frame that makes the face read more square than the bare jaw did. Soften, in other words, is bounded: a round frame softens the jaw reading exactly as long as it holds the width line and the decentration ceiling, and the moment it trades either for the circle it stops softening and starts exaggerating.