Glasses for a Round Face: Four Numbers That Fight Each Other

Every number on the frame, in millimetres and inches

There are ten numbers behind any honest answer to eyeglasses for a round face, and three of them are on the frame. They are stamped on the inside of the temple arm in a fixed order — lens width, bridge width, temple length — written either with dashes or with a small box symbol between the first two, so that 54-18-140 and 54□18-140 are the same frame. Two more are on you and nowhere else: your cheekbone width, which is the line the frame has to reach, and your pupillary distance, which comes from the refraction and never from a catalogue. The remaining five are computed, and they are the five that almost every round-face styling page leaves out, which is why those pages can recommend a frame that no laboratory will agree to make.

The table gives all ten, in millimetres with inches alongside. Where two published sources disagree, both figures are in the footnote rather than an average of them.

Published, and dated in the table further down: cheekbone width from a 2022 anthropometric dataset, men 142.62 ± 6.22 mm (5.61 ± 0.24 in) and women 133.77 ± 5.56 mm (5.27 ± 0.22 in), the range here being −1 SD to +1 SD across both; lens width, bridge width and temple length from a consumer eye-health reference published January 2019 and updated April 2026; stock blank diameters from a lens supplier, September 2026. PD is where the sources part company — a hospital health article published April 2026 gives 50 – 70 mm with an average of 63, an eye-health reference updated August 2025 gives 60 – 65 mm as the adult average, and neither is wrong about the population it measured. Computed, not published: Frame PD is the sum of two stamped numbers; ED = √(A² + B²) for a boxed rectangle, which is this page's own upper bound and slightly overstates a real lens because real lenses have rounded corners, while for a true circle ED = A. This page's own estimates, and labelled as such: the lens-depth range of 30 – 45 mm, taken from the per-style depths a fitting guide published in July 2026, and the total-width band, which is a fitting rule rather than a measured range — the published ceilings on it disagree, and the total-width section below sets them out. This page carries print styles; both tables come out complete on A4 or Letter.

Measuring five of them yourself, with a soft tape and a bank card

Five of those ten numbers you can get in about five minutes without an optician, and the tool that makes the fifth one possible is in your wallet. Every payment card on earth is cut to the same standard — ISO/IEC 7810 ID-1, 85.60 mm long by 53.98 mm tall, which is 3.37 in by 2.13 in — so a card is the only calibrated length most people own. It is too short to span a face on its own, but it is long enough to convert a photograph or a mirror mark into a number: a 138 mm (5.43 in) cheekbone width is 1.61 cards, and a face width that comes out at 1.5 cards is 128 mm (5.04 in).

For the PD itself the mirror method is the published one. Stand about 8 in (203 mm) from a mirror, hold a millimetre ruler horizontally against your face at brow height, close one eye, line the zero mark up with the centre of the other pupil, then switch eyes and read the mark that now sits at the centre of the first. That reading is your binocular PD. It is a fallback. If you have worn glasses before, your last prescription carries the number, and the optician who made them has it on file — ask, because a mirror reading is not as repeatable as a pupillometer and this page's arithmetic is sensitive to a millimetre.

The two numbers nobody tells you to write down are the total width and the lens depth of a frame that already fits you. Neither is stamped. Total width you get by laying a ruler straight across the front, from the outermost point of one end piece to the outermost point of the other. Lens depth you get by measuring the opening vertically at its tallest. On a round face the depth matters more than usual and is missing from the stamp more often than not, so if you are ordering online, ask the shop to measure it rather than assuming the shape name tells you.

• Take the cheekbone width first, with hair pushed fully back. Every other number on this page is conditional on it.
• Take face length too, hairline to chin. The bridge section further down needs the ratio between the two.
• Get PD from your last prescription if one exists. Use the mirror method only to fill a gap.
• Write it down as distance PD. Near PD is 3 to 4 mm (0.12 to 0.16 in) smaller in total, so a 62 mm (2.44 in) distance PD becomes 58 – 59 mm (2.28 – 2.32 in), and a frame chosen on the near figure will sit too narrow for distance wear.
• Read the three stamped numbers off a frame that already fits, not off one you like the look of.
• Measure that frame's total width and its lens depth with a ruler, and write both down. Neither is on the arm.
• Subtract twice the lens width and once the bridge width from that total width. The remainder is that frame's end-piece allowance, summed across both sides — the number the end-piece section below compares against four published medians.

Total width, and what a round face loses when the frame comes up short

Total width is not a fourth stamped number hiding somewhere. It is a sum, and the sum has a useful second form. Call the end-piece allowance R, added across both sides of the front. Then total width T = 2A + DBL + R. But Frame PD is A + DBL, so the same width can be written T = A + Frame PD + R. Check it on a 54-18 frame carrying the published acetate allowance of 20 mm (0.79 in): 2(54) + 18 + 20 = 146 mm, and 54 + 72 + 20 = 146 mm (5.75 in). Both forms agree, and the second is the one that matters, because it splits the width into the part your prescription pays for — Frame PD, which drives the decentration and the lens blank — and the part it does not.

Why the total is the number that governs this face specifically: on a round outline the three widths run close together and the cheekbone is the widest of them, so a frame that ends inside the cheekbone leaves bare skin as the widest line on the face. That is the exact reading a round face is trying to lose, and it is why small frames get warned against on this shape by pages that give no other reason. Three things go wrong at once when the frame is short. The cheekbone protrudes past the frame edge, so the face reads wider and the side contour reads more prominently than it measures. The temples have to splay outward to reach the ears, which pinches at the side of the head and tilts the front forward. And the decentration runs negative — Frame PD below your PD — which means the optical centres have to be moved outward rather than inward, and the minimum-blank formula below was written for the inward case.

How much wider than the face is too wide is where the published advice actually disagrees, and the disagreement is worth seeing because it hides the real ceiling. A marketplace fitting guide from June 2026 says the frame width should come within 2 to 4 mm (0.08 to 0.16 in) of the face. A frame-fitting software guide from July 2026 says equal to your face width or a few millimetres wider. Neither derives its number; each offers it as a fitting rule. Both agree on the floor, and the floor is the only part of this that is not in dispute: never narrower than the face. They differ on the ceiling, and the ceiling is not a styling choice at all. It is set by your pupillary distance, through the blank-size formula, and neither guide says so. Worked out for the face this page uses — 138 mm (5.43 in) across the cheekbones, a 62 mm (2.44 in) PD — it lands at 7 mm (0.28 in) past the cheekbone, and it does not land on a round number. A 53-19 frame carrying the published acetate allowance totals 145 mm (5.71 in), sits 7 mm past, and needs a blank of exactly 75.0 mm (2.95 in), which is the largest diameter anybody stocks. Add one millimetre of lens width and the total goes to 146 mm (5.75 in), 8 mm past, and the blank it needs is 75.8 mm (2.98 in) — a size that does not exist. That 7 mm is this page's own arithmetic from published inputs, not a published figure, and it moves when your PD moves.

The collision: width wants bigger, the lens wants smaller

Four constraints bear on the same frame. Three are physical and one is styling. Width says total width must reach the cheekbone — on a 138 mm (5.43 in) face, T ≥ 138. Decentration says (Frame PD − PD) / 2 must stay under 5 mm (0.20 in) per eye, published May 2026, which at a 62 mm (2.44 in) PD caps Frame PD at 72, and at this page's tighter 3 mm (0.12 in) target caps it at 68. The blank says MBS = ED + (Frame PD − PD) + 2 mm has to fit a stock diameter, of which the largest commonly held is 75 mm (2.95 in). And depth says the lens should stay under about 40 mm (1.57 in), published July 2026, because this is a short face and a deep lens covers what little mid-face there is — a styling limit that turns out to be a physical one as well, since B is the other term under the square root in ED.

Solve the three physical constraints at once and they collapse into a single bound on the lens. Frame PD ≤ 72 puts the decentration term at 10 mm (0.39 in), so MBS ≤ 75 requires ED ≤ 75 − 10 − 2 = 63 mm (2.48 in). For a boxed rectangle ED = √(A² + B²), which means every millimetre of depth is paid for in lens width. At B = 32 mm (1.26 in), A can run to 54.3 mm (2.14 in). At B = 34 (1.34 in), A ≤ 53.0 (2.09 in). At B = 36 (1.42 in), A ≤ 51.7 (2.04 in). At B = 38 (1.50 in), A ≤ 50.2 (1.98 in). At B = 40 (1.57 in) — the published depth ceiling itself — A ≤ 48.7 mm (1.92 in). On a round face, depth and width are not two constraints. They are one, and the exchange rate between them is the diagonal.

The width floor then decides whether a frame that survives all of that is wide enough, and this is where the end-piece enters. With A at its ceiling and DBL as wide as Frame PD ≤ 72 permits, T = 2A + DBL + R ≥ 138 sets a minimum allowance R of 13 mm (0.51 in) at B = 34, 15 mm (0.59 in) at B = 36 and 16 mm (0.63 in) at B = 38. Four published medians for R sit on the other side of that comparison; the next section takes them on. First, the same face fitted with seven frames, each carrying the published allowance for its material.

Every cell is arithmetic, not measurement: total width = 2A + DBL + R; decentration = (Frame PD − PD) / 2; ED = √(A² + B²) for a boxed rectangle and = A for a true circle; MBS = ED + (Frame PD − PD) + 2 mm. The allowance R in each row is the published median for that material — acetate 20 mm (0.79 in), plastic 16 (0.63 in), titanium 15 (0.59 in), metal 13 (0.51 in) — measured across 3,942 retailer frame records and published in August 2026; an individual frame sits either side of its material's median, and the last step in the measuring section is how to find out where yours sits. “Past” in the second column means past the cheekbone width, not past any published ceiling; the ceiling derived in the total-width section above is 7 mm (0.28 in) past for this face, which rows five, six and seven exceed — row five by a single millimetre of lens width, and that millimetre is what takes its blank from exactly 75.0 mm (2.95 in) to 75.8 mm (2.98 in). Row two is modelled as a true circle, which is its worst case: flattening the top rim, as the published exception describes, lowers B and therefore lowers both ED and MBS, so a flat-top version of that frame is strictly easier to glaze than the row shows.

The lever nobody prices: the end piece

Read the fourth column down and the boundary between wearable and not is one millimetre of lens width. 53-19 on an acetate front needs a 75.0 mm (2.95 in) blank, and a 75 mm blank exists. 54-18 on the same front needs 75.8 mm (2.98 in), and there is nothing to cut it from — 0.8 mm (0.03 in) of glass that no stocked diameter covers. Rows two, three and four clear all three physical constraints. Rows five, six and seven clear none of them. Nothing about the shape of the lens decided any of it.

Width was never the binding constraint, which is the part that runs against the keyword. Five of the seven rows reach a 138 mm (5.43 in) cheekbone; only row one comes up short, at 125 mm (4.92 in), which is 13 mm (0.51 in) short. The rows that fail, fail on the lens, and they carry the widest lenses in the set rather than the roundest — 57 mm (2.24 in) and 60 mm (2.36 in), whose boxed diagonals come to 68.5 mm (2.70 in) and 72.1 mm (2.84 in) and whose blank requirements come to 83.5 mm (3.29 in) and 86.1 mm (3.39 in) against a 75 mm (2.95 in) maximum. Row two is the sharpest counterexample in the table: a true circle, B/A 1.00, the ratio the whole shape argument turns out to be about — and it needs only a 64.0 mm (2.52 in) blank, because a circle's effective diameter is its width and not its diagonal. A large round lens is the cheapest shape on this list to glaze. It is the small one that is not.

The allowance is the one number in this calculation that the published record disagrees about by a factor of three, and the disagreement is recent. A fitting guide published in July 2026 puts the end piece at 3 mm (0.12 in) per side, which is 6 mm (0.24 in) across the frame. A retailer's rule of thumb puts the two outer rims at 5 to 8 mm (0.20 to 0.31 in) in total. A frame-specification comparison published in August 2026 assumes a flat 10 mm (0.39 in) in its own formula and then does what none of the others did, which is test the assumption against measurement: median remainder by material, over 3,942 retailer frame records — acetate 20 mm (0.79 in), plastic 16 mm (0.63 in), titanium 15 mm (0.59 in), metal 13 mm (0.51 in). Those run two to three times the assumed figures. The same page is explicit that it cannot establish how any individual frame's width was obtained, which is why the table above labels the material on every row rather than pretending to one number, and why the last step of the measuring section is how to find your own frame's.

Run the algebra the other way and the medians turn into a criterion you can shop against. Total width has to reach the cheekbone, so the allowance has to satisfy R ≥ face width − lens width − PD − 2 × decentration ceiling. Combine that with the blank ceiling — MBS at or under 75 mm means ED at or under 63 mm (2.48 in), which on a boxed rectangle caps the lens at 53 mm (2.09 in) for a 34 mm (1.34 in) depth, 51 mm (2.01 in) for 36 mm (1.42 in), 50 mm (1.97 in) for 38 mm (1.50 in) — and the required allowance comes to 13 mm (0.51 in) at a 34 mm depth, 15 mm (0.59 in) at 36, 16 mm (0.63 in) at 38 and 18 mm (0.71 in) at 40 mm (1.57 in). Set the medians against those. Acetate clears every depth up to the 40 mm ceiling. Plastic clears to 38. Titanium to 36. Metal clears at 34 mm only, and only with a lens no wider than 53 mm — the narrowest margin in the set, on the material that looks thinnest on the face. At a 142 mm (5.59 in) cheekbone and a 62 mm (2.44 in) PD the arithmetic gets worse rather than better: the requirement rises to 17 mm (0.67 in) at the 5 mm ceiling and 21 mm (0.83 in) at this page's 3 mm target, and only acetate reaches either.

The margins differ because the three levers are not priced alike. One extra millimetre of lens width buys 2 mm (0.08 in) of total width, costs 0.5 mm (0.02 in) of decentration, and raises the diagonal — so it tightens the blank at four times the rate it loosens the width. One extra millimetre of bridge buys 1 mm of width for the same 0.5 mm of decentration: two to one. One extra millimetre of end piece buys 1 mm of total width and costs nothing at all, because it sits outside the lens and Frame PD never moves. Rim is free width. That is an optical result rather than a styling opinion, and it derives three pieces of round-face advice that every page gives and none of them justifies: that a bold frame does more for a wide round face than a delicate one, that rimless and pale and transparent fronts do very little, and that what this shape should avoid is the small and the flimsy rather than the round.

How round is too round, as a ratio rather than a slogan

The keyword comes with a premise and the premise does not survive contact with the published record. Half the advice says a round face must avoid round frames. A value optical chain's fitting page goes the other way and says the shape to avoid on a rounded face is square, because square puts rounded features out of balance — while a brand page four months later puts square first, and an optometry practice a year before that asks for rectangular and angular frames to make the face look sharper and longer. Those three cannot be describing the same object. They are not.

The instrument that separates them is B/A: lens depth divided by lens width. A true circle is 1.00. A true square is also 1.00. The shape-name argument is therefore not about round versus square at all — it is about how close B/A sits to one, and the names on the shelf do not track it. Take the per-style targets a fitting guide published in July 2026 and divide. Rectangle, A 52 – 56 mm (2.05 – 2.20 in) by B 32 – 38 mm (1.26 – 1.50 in), is B/A 0.57 – 0.73. Square, 50 – 54 by 36 – 40 mm (1.97 – 2.13 by 1.42 – 1.57 in), is 0.67 – 0.80. Wayfarer 0.69 – 0.81. Browline 0.67 – 0.80. Cat-eye 0.63 – 0.76. Geometric, hexagon and octagon, 0.63 – 0.80. Every frame that page recommends lands in one band, 0.57 to 0.81. The single row it marks as the mistake — small round wire, 44 – 48 by 40 – 44 mm (1.73 – 1.89 by 1.57 – 1.73 in) — lands at 0.83 to 1.00.

So “square” in a fitting guide means B/A near 0.7, and the page telling you to avoid square is avoiding B/A near 1.0 and calling it square. Both camps are right about the geometry and wrong about the label, which is the ordinary explanation for advice that reads as contradiction. The best frames for a round face, on this measure, are the ones between 0.57 and 0.81, and the best glasses frames for a round face are not a family of shapes but a band of ratios that six differently-named families all happen to sit inside.

Stated as conditions rather than a slogan, with the numbers attached, there are four — and it matters that three of them are physical and one is not. Physical: total width reaches the cheekbone; decentration stays at or under 5 mm (0.20 in); and the minimum blank size the frame implies falls inside a stocked diameter, 75 mm (2.95 in) at the top of the range. Styling: B/A at or under 0.80. The four are independent, and row two of the table above proves it. That row is a true 52 mm (2.05 in) circle at B/A 1.00, and it satisfies all three physical conditions while failing the fourth outright: 144 mm (5.67 in) of total width against a 138 mm (5.43 in) cheekbone, decentration exactly at 5.0 mm, and a 64.0 mm (2.52 in) blank that a 65 mm (2.56 in) diameter covers. It is also the published exception to the avoid-round rule — a large round lens, 52 mm or more, with a flat top rim or a strong brow bar — and the exception is not an exception to the arithmetic. Flatten the top and B drops without A moving, so B/A drops with it and ED and MBS drop after that. The same frame then passes all four, and passes them more comfortably than the row shows, because the row models it as a true circle, which for blank size is its worst case.

It fails, and fails for a reason that has nothing to do with roundness, when B/A is high and the lens is small. Row one of the table is the small round wire every guide names as the mistake: decentration a comfortable 2.0 mm (0.08 in), a trivial 52.0 mm (2.05 in) blank — and 125 mm (4.92 in) of total width against a 138 mm (5.43 in) face, which is 13 mm (0.51 in) short. It is not too round. It is too small, and a small square frame of the same width would fail identically.

Two more bounds, both with numbers. Past eight sides a polygon reads as a circle again — published July 2026 — so “geometric” is not an escape from the ratio; a 50 × 40 mm (1.97 × 1.57 in) octagon is B/A 0.80 and sits at the very top of the safe band. And the depth ceiling is a constraint separate from the ratio, and the one this face hits first: under about 40 mm (1.57 in). On a high-minus prescription the same ceiling bites for a second reason, which is a derivation rather than a published rule — a deeper lens carries more of the thick edge into view, and where there is astigmatism with its axis near horizontal the thick edge runs vertically, top and bottom, which is precisely where a deep lens puts it. If that is your prescription, choose a shallower frame and let the ratio look after itself.

The bridge sits high on this face, and the reading that says why

Everything so far has been horizontal. There is one vertical lever, and it is specific to this outline rather than shared with the other six. A round face is defined by a low face-length-to-cheekbone ratio — this site classifies on that reading, and it runs from about 1.05 at the round end to about 1.65 at the long end. A low ratio means a short mid-face: not much distance between the brow line and the base of the nose.

The bridge bar is the frame's only horizontal line at mid-face height, so where it sits sets how that short stretch is divided. The conclusion — and this is a derivation from the ratio, not a published rule — is that on a short mid-face the bar wants to sit high, because a low bar on a short stretch leaves two segments that are both too small to read as anything, while a high bar puts the line near the brow and leaves the whole of the mid-face below it as one unbroken span. On a long mid-face the instruction inverts. Which is to say that the round-face-specific recommendation is a high bridge: a browline, a clubmaster, a keyhole bridge that sits above the widest point of the nose. That is the same conclusion the shape guides reach when they list browline and cat-eye for this face, arrived at from the measurement instead of from the silhouette.

The test is in the mirror, not in a table. Put the frame on and find the line from your brow to the base of your nose. If the bar sits at or above the top of that line, the frame is doing the short-mid-face job. If it sits low across the middle of it, the same frame on a longer face would be right and on yours is cutting the face in half.

Two further checks that are about soft tissue rather than bone, and that a frame can fail while every number in the table passes. The lower rim must not touch the top of the cheek when you smile. And the widest point of the front should sit level with the cheekbone rather than above or below it — level, because that is what makes the frame the widest line on the face instead of the cheekbone. Both are touch tests, both take ten seconds, and neither is on any spec sheet. Bridge width follows the nose rather than the face: a wide nose bridge or wide nostrils want a narrower DBL and end-pieces that turn inward, a long nose wants the reverse.

Round-face glasses advice, by the date it was published

Seventeen positions, one per row, each labelled with the date it was published. Source names are withheld as a matter of house style on these tables; the kind of source is given instead, and the dates are publication dates rather than the dates anything was written. The load-bearing claim is not any single row's attribution — it is the difference between the third column and the fourth, read top to bottom.

One position per row, in publication-date order; the undated row is last because there is nothing to order it by. The third column is what each source actually published as a number, converted to inches at 25.4 mm to the inch, and it is never this page's own estimate. Where a source gave a worked example the example is reproduced rather than paraphrased. Where a page was updated after publication both dates are given, and the row is ordered by the original publication date.

What swung, what held still, and the one number that moved

Read the third column down. It is not empty — it runs from January 2019 to September 2026 — but not one number in it was written for a round face. Every figure there is about frames and pupils in general. What the column does carry is seven years and eight months of arithmetic that has barely moved. Lens width 40 – 60 mm (1.57 – 2.36 in) in January 2019; lens width 40 – 60 mm in August 2023; the same ranges again in June 2026. Pupillary distance moved from “60 to 65” to “50 to 70, average 63”, which is a widening of the stated range and not a change in the thing measured. In that window exactly one number in that column has been contradicted by a better one.

It is the end-piece allowance, and it moved in the closing months of the window. In July 2026 a fitting guide published it as 3 mm (0.12 in) per side. In August 2026 a frame-specification comparison published it as a median measured over 3,942 retailer frame records, broken out by material: acetate 20 mm (0.79 in) across the frame, plastic 16 mm (0.63 in), titanium 15 mm (0.59 in), metal 13 mm (0.51 in) — which is 6.5 to 10 mm (0.26 to 0.39 in) per side. The two differ by a factor of two to three. One is an assumption and the other is a measurement, and the measurement is the one that changes what fits. At a 138 mm (5.43 in) cheekbone, a 62 mm (2.44 in) pupillary distance and a 53 mm (2.09 in) lens held to the 5 mm decentration ceiling, the bridge can be at most 19 mm (0.75 in); on a 6 mm allowance the frame totals 131 mm (5.16 in) and stops 7 mm (0.28 in) short of the cheekbone, while on a 20 mm allowance the same frame totals 145 mm (5.71 in) and carries past it. Same face, same prescription, same lens, same bridge. The only retraction-grade change in the whole record is in the one term nobody prices.

Read the fourth column down and it swings three times in twenty-one months. Angular only, October 2024. Avoid square, November 2025. Square first, April and July 2026. The ratio section above resolves the swing arithmetically: the July 2026 square is B/A 0.67 – 0.80 and the November 2025 square is B/A near 1.00, so the two pages are not contradicting each other, they are using one word for two different ratios. That is worth knowing, because it tells you which half of the disagreement is real. None of it is.

The genuinely new content in the window is that laboratory-side numbers reached consumer-facing pages at all. Minimum blank size appears in a consumer technical guide in November 2025. The boxing system and the effective diameter in June 2026. Stock blank diameters in September 2026. Before November 2025 nothing in this set constrained what could actually be manufactured, which is the honest reason the older shape advice reads as pure taste — when it was published there was nothing to check it against and, for a round face, there still is not, because the two rows that mention a lens blank mention no face shape at all, and of the four rows that name a shape exactly one carries a millimetre.

That gap is what this page exists to close, and it is also the answer to the question the keyword is really asking. The best glasses for a round face are not a shape. They are a frame that clears four constraints at once, and the four divide three to one. Three are physical and no amount of taste argues you out of them: total width at or above your cheekbone width; decentration at or under 5 mm (0.20 in) per eye and preferably under 3 mm (0.12 in); and a minimum blank size inside a diameter somebody stocks, which on a boxed rectangle caps the lens at 53 mm (2.09 in) for a 34 mm (1.34 in) depth and lower as the depth rises. One is styling, and it is the only one of the four that is a matter of taste: B/A at or under 0.80. The best eyeglasses for a round face are whichever frames satisfy all four — and there will be more than one, and they will not all be called the same thing.