Face Form Finder

Round-Faced Celebrities, Measured

Twenty-eight people whose names keep coming up when the question is which celebrities have a round face. The index (face length ÷ cheekbone width) × 100, on one straight-on photo each, runs from 106.23 to 142.52, median 120.06. The last four in that order read 127.68, 132.04, 134.04 and 142.52. Being last in this order is not a verdict on a face — it is one index, on one photograph, on one day.

Rows run in that order, smallest index first. The median of 120.06 falls between row 14 and row 15 and is marked in the table. Every reading in the table was measured on 2026-10-20. — Download the CSV

The lists behind the names, dated

Every name in the table above came off a list that was put online, and every list carries its own date and its own count. This is the run-down, oldest first, with the number of rows this page took from each. The dates are the dates those pages put on themselves; they are not the dates the photographs were taken, and a page that was updated later can still be holding a much older list.

Four of those eight pages account for the 28 rows in this table: 15 from the January 2023 list, 8 from August 2024, 3 from January 2025 and 2 from September 2026. The other four pages contributed no rows to this table. Those counts describe this table's selection, not how many usable photographs each page contains.

What the 28 readings say

The lowest six index values in the set are Quenlin Blackwell 106.23, Christina Ricci 108.84, Gemma Chan 110.60, Megan Thee Stallion 114.29, Drew Barrymore 114.75 and Halle Bailey 114.78. The highest four are Dakota Johnson 127.68, Kelly Osbourne 132.04, Queen Latifah 134.04 and Michelle Williams 142.52. Between the two ends sits 36.29 points of spread, with the median at 120.06.

The last four are the four largest index values in this set, in this order, on these photographs. That is the whole of what the order says. It does not sort faces into named shapes and it does not say what any of these people is or is not. A face can sit at the top of this column and still be called round by everyone who writes about it, and a face at the bottom can be called the same thing; the column measures a photograph, not a word.

Two rows sit either side of the median: Vidya Balan at 120.00 and Jennifer Lawrence at 120.12.

What the comment threads add

The following examples come from comments returned for five discussions. Some comments were removed or unavailable; the returned text is not a complete record of every reply. These examples describe individual opinions, not measured effects or a survey of face shapes.

Advice about hair and reference pictures

Face-framing advice appeared in three of these discussions: about a short cut, natural hair and an emo cut. One person also suggested bringing a stylist pictures matched to both face shape and hair type. These are suggestions from commenters, not a test showing which haircut works for a particular person.

Different labels for the same face

In one discussion, the poster asked whether their face was round or heart-shaped. The returned replies included oval, classic oval, diamond, heart, an oval-and-heart mix, and round. The labels are retained here as written; classic oval and oval are not counted as separate anatomical categories. The replies disagree, but do not establish a correct classification.

In two other discussions, a poster described their face as round while replies described it as oval. In the natural-hair discussion, a commenter called the attention to face shape overly nitpicky. That comment was not from the round-or-heart discussion above.

What these accounts cannot establish

One person said they had often been told their face was round, especially with their hair tied up. This is a personal account, not a paired measurement of hair up versus hair down. The table above does not record hairstyle state and cannot quantify that effect.

Two separately collected comments expressed concern that facial slimming procedures change familiar examples of round faces. They do not establish how common those procedures are, when they became common, or whether the dated lists above precede that change.

How each number was produced

The three widths and one face-length reading are detector outputs from one photograph, not group averages. They use distances between four fixed landmark pairs, expressed in image pixels; the saved values are rounded display readings, not direct anatomical measurements. Here is the run, step by step, so you can repeat it with your own photo:

1. Open the detector on this site in a browser. It runs on your own device; nothing is uploaded.
2. Load one straight-on photograph with a single face in it. The run below used one photo per person.
3. Let the detector place its points, then read the four distances it prints: forehead, cheekbone, jaw and face length.
4. Divide each of the three widths by the face length. That is what the first four data columns are, to two decimals.
5. Divide face length by cheekbone width and multiply by 100. That is the seventh column, to two decimals, and it is a calculated figure built from the two saved, integer-rounded pixel readings.
6. Read the jaw width against the cheekbone width. That ratio is the second number in the basis column.

The detector uses four fixed landmark pairs for the three labelled widths and the face-length reading; it does not search for the widest span or locate the actual hairline. The saved pixel counts behind every row are rounded display readings, and each row's arithmetic was recomputed from those counts.

The table contains 28 successful detector readings. The listing pages did not state image licences, and the CSV does not supply capture dates. Capture-date information is recorded for one separate Commons test photograph, not established for the photographs in this table. Listing-page update dates are not photo capture dates. Two names that would otherwise belong are missing for a different reason: one photograph came back too small to place points on, and one came back with no face the detector could isolate. Neither was filled in with a guess.

A worked unit-conversion example

These are invented example inputs for arithmetic, not measurements of a person or rows from the table. A ratio requires both lengths in the same unit; it does not require converting millimetres to inches.

One inch is exactly 25.4 millimetres by definition. An example length of 22.5 centimetres is 225 millimetres, or 225 ÷ 25.4 = 8.8582… inches, rounded to 8.86. An example width of 14 centimetres is 140 millimetres, or 140 ÷ 25.4 = 5.5118… inches, rounded to 5.51.

The example index is 225 ÷ 140 × 100 = 160.71, rounded to two decimals. This exceeds the table's maximum of 142.52. Its width relative to length is smaller: 140 ÷ 225 = 0.6222…, compared with approximately 100 ÷ 142.52 = 0.7017 for the table's highest-index row. This arithmetic does not establish whether those example dimensions are common.

Back-converting the rounded length gives 8.86 × 25.4 = 225.044 millimetres, displayed as 225.04. Keep unrounded values through the calculation and round the final displayed result. A tape measurement and a detector's image-pixel distance are different measurements; this example does not calibrate one against the other.

Three limits of these photograph readings

The tests below report detector outputs after rescaling or rotating one existing image. They do not measure actual camera pitch, lens effects or changes of expression. The results come from the same Selena Gomez photograph; they are not rates across people.

1. Image scaling and in-plane rotation — measured

The test used one photograph from the table, the Selena Gomez row, whose index is 120.85. Squeezing the frame 5% vertically moved the index to 115.02 (-5.83). Stretching it 5% vertically moved it to 128.77 (+7.92). These were image-height edits, not a calibrated five-degree pitch test; real camera pitch was not measured.

Rotating the same photograph 5 degrees in the picture plane returned a +0.00 difference in the index calculated from the displayed integer pixel readings. This does not establish an unrounded zero or a rounding-free result. Unrounded landmark coordinates were not retained, so the result is limited to the saved display precision.

2. Lens focal length — not measured

Lens focal-length effects were not measured. No focal-length comparison of the same face was made here, so the image-rescaling tests do not quantify perspective changes from a real camera or lens. They establish only the detector outputs for the edited image.

3. Horizontal image scaling — measured; expression — not measured

The same photograph was squeezed 3% and then 5% across its width. The 3% squeeze moved the index from 120.85 to 125.00 (+4.15); the 5% squeeze moved it to 128.57 (+7.72). In the 5% test, the saved face-length reading also changed from 255 to 261 pixels, so the result cannot be described as a cheekbone-only change with face length held constant.

Expression was not measured. These records contain no paired expression test, so they do not establish a direction or size of an expression effect on these readings.

The scaling results describe edits to one image and a detector rerun. They do not establish that differences between two table rows are caused by camera angle, retouching or differences between faces.

What this table does not do

It does not rank faces, and the order is by a ratio rather than by anything about how a face looks. It does not assign a face shape, and no row carries a shape name. It does not settle which of these people has a round face, because a single photograph cannot settle that and because the words round, oval and square overlap across most of the range these rows sit in.

What it does do is put a number next to each name, so that the next time a list tells you a face is round, you can see what that face measured and where the figure came from.

The four readings that build a face shape from the brow down are explained in Face Shape Chart, and the hairstyles that follow from them are in Bangs for a Round Face and Bob Haircut for a Round Face.