Most lists of lip shapes arrive as a set of names with a sentence for each: full, thin, wide, round, heart-shaped. The names are useful shorthand, but they are not one classification. One mouth can be wide, bottom-heavy and bow-shaped at the same time, because those words answer different questions. The list looks like a menu of alternatives only because the questions are folded into a single line.
This page unfolds them. It starts with four readings that a ruler can take and a 2024 study measured on 400 women, with millimetres and inches printed side by side. It then sorts the common labels by the family they belong to, records what the dated guides actually publish, explains where a label stops being useful, and ends with what the readings change for liner, gloss and photographs. The order is: the readings, the families, how to take them, the ratios and the printed bands, the dated sources, what changed, what a shape cannot tell you, and what it changes in practice.
Four readings, and the mean span they fall in
The four readings are all distances. Mouth width runs from one cheilion, the corner where the upper and lower lips meet, to the other. Upper vermilion height runs from labiale superius, the lowest point of the upper lip's wet border at the midline, down to stomion, the line where the closed lips meet. Lower vermilion height runs from stomion down to labiale inferius, the midpoint of the lower lip's wet border. Cupid's-bow height is the vertical distance from the central dip of that upper border to either peak, the two points called crista philtri.
These four were measured on 400 professional female photograph models, Caucasian, aged 18 to 39, in a 2024 photogrammetric study. A semi-automatic analysis traced the landmarks on frontal photographs. The table prints the mean and the standard deviation for each reading. The sample is a set of professional models, not a cross-section of the population, and a mean is a centre, not a target.
Four lip readings from 400 Caucasian female models aged 18–39, measured by semi-automatic photogrammetry in a study published online in 2024. The table prints the group mean and one standard deviation. Inches are the millimetres divided by 25.4 and rounded separately. The sample is a set of professional models, not a population norm.
| Reading | Where it is taken | Mean (SD) | What it separates |
|---|---|---|---|
| Mouth width | Cheilion to cheilion, corner to corner | 48.1 mm (1.89 in), SD 3.3 mm | The wide and narrow ends of the outline |
| Upper vermilion height | Labiale superius to stomion, at the midline | 6.5 mm (0.25 in), SD 1.4 mm | A thin upper lip from a full one |
| Lower vermilion height | Stomion to labiale inferius, at the midline | 11.6 mm (0.46 in), SD 1.5 mm | A thin lower lip from a full one |
| Cupid's-bow height | Central dip to either peak of the upper border | 8.2 mm (0.32 in), SD 1.3 mm | A flat bow from a defined one |
A standard deviation is the spread of individual readings around the mean, so the four means above are not the edges of the sample. The same study measured the distance between the two bow peaks at 12.3 mm (0.48 in) and lip obliquity, the angle of the mouth line against horizontal, at 1.05° with a standard deviation of 0.79°. A separate study of 156 Black South African women aged 18–25 reported a within-examiner Dahlberg coefficient of 0.102 mm for upper vermilion height and 0.093 mm for lower vermilion height. That coefficient is an estimate of repeat-measurement error, not the spread between people; it is roughly a tenth of a millimetre, far smaller than either standard deviation above. The samples, methods and dates behind these numbers are set out in the dated table in section five.
No account · No ad pixels · No photo upload
Detection runs on-device with MediaPipe. Nothing is uploaded — no account, no email, no storage.
The labels, sorted by the question each one answers
The two 2026 guides read for this page both list nine labels: full, thin, wide, round, heart-shaped, bow-shaped, top-heavy, bottom-heavy and downturned. The list looks like nine alternatives. It is closer to nine sentences about six different axes, and the axes do not exclude one another. A mouth can be wide, bottom-heavy, bow-shaped and slightly downturned in the same description, because width, balance, bow and corner direction are separate readings.
Thickness is the pair of vermilion heights. Balance is the ratio between them. Width is the distance corner to corner. Bow is the shape of the upper border around the two peaks. Outline is the curve of the whole vermilion border, which is what the word round usually points at. Corner direction is an angle, not a distance: a downturned mouth has corners that sit below the horizontal line through the centre of the mouth. The four readings in the first table cover the first four axes; they do not cover outline or corner direction.
The nine labels map onto those axes unevenly. Full and thin are thickness words. Top-heavy and bottom-heavy are balance words. Wide is the only width word in the two lists, and neither list names its opposite. Heart-shaped and bow-shaped both depend on the bow, while heart-shaped also leans on the outline above it. Round is an outline word. Downturned is the only corner word. Sorting them this way is more useful than memorising nine definitions, because it shows which labels can be checked with the four readings and which cannot.
The nine labels used by both 2026 guides, sorted by the question each one answers. The last column says whether the four readings in section one can settle the label under the rule one guide prints. That guide writes a rule for all nine labels, using its own scan of 18,630 faces, a ratio against cheekbone width, or a corner angle; the other guide prints the labels but no rule. The rules are that guide's, not findings of the 2023–24 measurement studies.
| Family | Labels in it | What it is read from | Can the four readings decide it? |
|---|---|---|---|
| Thickness | Full, thin | Lip height, upper plus lower, divided by mouth width | Yes: the guide reads full above about 0.43 and thin below about 0.33, both within its own scan |
| Balance | Top-heavy, bottom-heavy | Upper vermilion height divided by lower vermilion height | Yes: the guide reads top-heavy at 1.0 or above and bottom-heavy below 0.55 |
| Width | Wide | Mouth width divided by cheekbone width | No: the rule needs cheekbone width, which is not one of the four readings |
| Bow | Heart-shaped, bow-shaped | Bow depth divided by mouth width, plus the upper-to-lower ratio | Yes: the guide reads a pronounced bow above 3.5 percent of mouth width; heart-shaped adds an upper-to-lower ratio of at least 0.8 and fullness above its median |
| Outline | Round | Fullness plus mouth width against cheekbone width | Partly: fullness is covered, but the rule also needs cheekbone width |
| Corners | Downturned | Angle of the mouth corners against horizontal | No: the four readings are distances, not an angle |
The labels are listed as the two 2026 guides use them, not as a closed taxonomy. One of those guides prints a corner threshold of minus 5 degrees for a downturned mouth and bands of 6 to 9 mm for the upper vermilion and 8 to 11 mm for the lower; it also reports its own scan of 18,630 faces, measured on its site on 4 September 2026, as percentile tables. From those ratios, mouth width against cheekbone width and the corner angle, it writes a rule for each of the nine labels; the table above records them. That scan covers ratios, not the two millimetre bands, which the guide attributes to population measurements without naming the sample. The other guide prints a single average width, about 51 mm, attributed to published anthropometric studies, plus a sample output from its detector. Neither guide reports a standard deviation or a repeat-measurement error, so a band printed without an error term cannot be read as a range that holds across populations.
How to take the four readings with a ruler and a mirror
Work in even light, facing a mirror, with the lips lightly closed and the face relaxed. For mouth width, hold a millimetre ruler level, place the zero at one corner and read the other corner without pressing into the skin. For upper vermilion height, find the lowest point of the upper wet border at the midline and measure straight down to the line where the lips meet. For lower vermilion height, measure from that same lip line down to the midpoint of the lower wet border. For Cupid's-bow height, find the two peaks and the central dip, then measure vertically from the dip to one peak. Take each reading twice, on two days if possible, and keep the average.
The study methods are a useful scale for what precision looks like. The Indonesian study photographed participants from one metre with a ruler in the frame and measured the images in software; it used a single investigator and does not report a repeatability figure. The South African study did report one, and its Dahlberg coefficients for the two vermilion heights were 0.102 mm and 0.093 mm. Those are small numbers, but they come from a research workflow with a calibrated image, not from a hand-held ruler in a bathroom mirror. A home reading should be treated as a reading with a method attached, not as a laboratory value.
The arithmetic shows which reading is most fragile. One millimetre is about 15 percent of a 6.5 mm upper vermilion, about 9 percent of an 11.6 mm lower vermilion and about 2 percent of a 48 mm mouth. A one-millimetre slip on the upper lip moves the upper-to-lower ratio from 0.56 to about 0.64, which is close to the study's own standard deviation of 0.10 for that ratio. A one-millimetre slip on the lower lip moves it to about 0.51. Both are possible with a ruler that is not held level, so the ratio should be read as a band, not as a decimal place.
What this means for you: write the method beside each number. A mouth width taken corner to corner is comparable with the first table; one taken from the wet line or from the centre of the philtrum is a different reading and should not be compared with it.
What the ratios decide, and what the printed bands leave open
Three ratios fall out of the 2024 means. Upper vermilion divided by lower vermilion is 0.56. The distance between the bow peaks divided by mouth width is 0.26. Total vermilion height divided by mouth width is 0.38. Ratios are the part that travels between face sizes, because the absolute width cancels; they are also the part that the two guides handle differently. One prints percentile bands from its own scan of 18,630 faces; the other prints an average width from published studies and a sample detector output.
The raw bands can be checked against the samples. One 2026 guide gives an average adult upper vermilion of 6 to 9 mm and a lower vermilion of 8 to 11 mm, attributed to population measurements without a named sample. The 2024 model sample has an upper mean of 6.47 mm and a lower mean of 11.64 mm: the upper sits inside the band and the lower sits above it. The Indonesian sample has female means of 6.59 and 10.02 mm and male means of 5.60 and 10.51 mm: three readings sit inside the bands and the male upper mean sits below. The South African sample has an upper mean of 11.0 mm and a lower mean of 13.2 mm: both sit above the bands. Of the eight sex-specific upper and lower means available across the three studies, four fall outside the printed bands.
The guide's own scan lands on the same centre for the ratio it measures itself. It puts the median of lip height divided by mouth width at 0.38, with the middle half between 0.33 and 0.43. The 2024 means give 18.1 mm of total vermilion on a 48.1 mm mouth, which is 0.38. The other guide's figure of about 51 mm for average adult lip width also sits within one standard deviation of the 2024 mean of 48.1 mm with a standard deviation of 3.3 mm. A large scan and a 400-model sample can agree on a centre while the spread around that centre stays wide, and neither number settles a label.
That is not a refutation of the guide's band. It is a warning about what a band without a named sample and an error term can mean. The three studies used different populations, different photographers and partly different landmark conventions, so some of the gap is method. But the direction is consistent enough to say that the band describes only part of the range in these samples, and cannot be used as a universal range. The comparison is also the reason this page prints the sample beside every mean rather than a single 'normal' line.
The preference study adds a second caution. Its 428 raters ranked a 1:2 upper-to-lower ratio as the most attractive of the variants they saw, using synthetic images of 20 white women. A 1:2 ratio is 0.5. The measured means here run from 0.56 in the model sample to 0.64 and 0.67 in the Indonesian sample and 0.84 in the South African sample. None of them is 0.5. A preference expressed by raters looking at manipulated images is not the same thing as the average ratio in a population, and neither one is a rule for an individual mouth.
What this means for you: the ratios make a description checkable, but they do not sort a mouth into a name. If a label does not say which axis it is using, it is not yet a description you can check.
Six dated sources, sorted by publication day
The table below records what each source actually publishes, in the order the pages were published. It is deliberately a mixed list: a preference study, three measurement studies and two guides. The point is not that one kind of source is better than another. It is that the labels, the bands and the measurements live in different documents, and a reader is left to join them.
Six dated sources, ascending by publication day: a 2017 preference study, three anthropometric studies from 2023–24 and two 2026 guides. Dates are the pages' own publication dates; for the journal articles the online date is used where the issue date falls later.
| Published | What it is | What it puts down | What it measures |
|---|---|---|---|
| July 2017 | Preference study in JAMA Facial Plastic Surgery; 428 raters, synthetic white female faces | A 1:2 upper-to-lower lip ratio ranked most attractive; no shape labels | Rater rankings of manipulated images, not a population mean |
| 28 April 2023 | Anthropometric study in Plastic and Reconstructive Surgery Global Open; 156 Black South African women aged 18–25 | Upper vermilion 11.0 mm, lower vermilion 13.2 mm; Dahlberg repeat error 0.102 and 0.093 mm | Means, medians, ranges and repeatability; no lip-shape labels |
| 24 December 2023 | Anthropometric study in JPRAS Open; 100 Indonesian adults aged 20–35, 47 men and 53 women | Mouth width 48.2 mm in men and 49.2 mm in women; upper vermilion 5.6 and 6.6 mm; lower 10.5 and 10.0 mm | Sex-specific means, ratios and ranges; no lip-shape labels |
| 28 August 2024 | Photogrammetric study in Aesthetic Plastic Surgery; 400 Caucasian female models aged 18–39 | Mouth width 48.1 mm; upper vermilion 6.5 mm; lower 11.6 mm; bow width 12.3 mm; bow height 8.2 mm | Means, standard deviations and ratios; no lip-shape labels |
| 5 September 2026 | Consumer guide, The Face Report | Nine labels; upper 6–9 mm and lower 8–11 mm attributed to population measurements; corner tilt below minus 5 degrees; own 18,630-face scan for percentile bands (fullness median 0.38) | Percentiles from its own scan; no standard deviation or repeat-measurement error; no named sample behind the millimetre band |
| 14 September 2026 | Consumer guide, Lip Shape Chart | Nine labels; average adult lip width about 51 mm from published studies; detector example 52 mm wide, 18 mm high, 1:1.6, bow angle 148° | No sample of its own, no standard deviation or repeat-measurement error; identification is by an outline detector |
The 2017 study's 428 raters were ranking variations of 20 synthetic faces; its 1:2 result is a preference, not a measurement of the population. The three anthropometric studies used different samples and partly different methods, so their means are printed side by side for comparison, not averaged into one figure. The 2026 dates are the pages' own publication dates. The 5 September guide reports its own scan of 18,630 faces for percentile bands but no standard deviation or repeat-measurement error, and does not name the sample behind its two millimetre bands; the 14 September guide reports no sample of its own and gives an average adult lip width of about 51 mm from published studies. The study that reports a repeatability figure is the South African one, not the source of the printed bands.
What changed, what did not
What changed between 2017 and 2026 is the vocabulary, not the measurement. The 2017 preference study worked with ratios and did not name a single lip shape. The 2023 and 2024 anthropometric studies measured lips and also named no shape. The two 2026 guides both name the same nine labels and turn them into identification checklists, which is the closest any of the six comes to a method. One of those guides prints millimetre bands, a degree threshold and percentile bands from its own scan; the other prints a single average width and a detector example.
What did not change is the separation between the two literatures. No measurement study in the set prints a lip-shape label. One 2026 guide reports a sample behind its percentile bands — 18,630 faces measured on its site — but neither guide publishes a standard deviation or a repeat-measurement error, and the millimetre band in the first is attributed to population measurements without a named sample. A guide can print a band without a named sample or an error term; a study can print a mean without the label it implies. The reader still has to join them, which is what the first and fourth tables above do.
The direction of travel is informative too. The 2026 guides agree on nine labels but disclose numbers at different levels: one reports percentile bands from its own scan and a millimetre band, the other gives a single average width and a detector example. Neither disclosure includes an error term, and a later date does not by itself make a guide more precise.
What this means for you: use the dated sources for their vocabulary and their questions, and use the measurement studies for means and error terms. The two kinds of source answer different questions, and the date alone does not tell you which one you are reading.
What lip shape does not tell you
A lip shape describes the vermilion border and the distances across it. It is not a health test. Nothing in the sources read for this page links a lip label to hydration, vitamin status, disease or wound healing, and this page does not make that link. It is not an ancestry test either: the three measurement samples are South African, Indonesian and Caucasian, and their means differ. A mean from one sample does not transfer to another person as a norm.
Nor does a label rank a face. The 2017 study is a preference study: its 428 raters were ranking synthetic images, and its most-preferred ratio was one option among the variants it showed. That is evidence about a group of raters, not a verdict on an individual mouth. The measured samples also overlap. In the Indonesian study, the male and female means differ by 0.5 to 1.1 mm on mouth width and the two vermilion heights, while the standard deviations for those readings are 1.4 to 6.3 mm. The two distributions overlap substantially, so a single reading cannot be used to infer sex, and a label cannot be used to rank one person against another.
The four readings also do not add up to a complete shape. They leave out the outline, the corner angle and the projection of the lips in profile. A frontal photograph can show the first two approximately, but projection needs a side view or a three-dimensional scan. A page that names a shape from a frontal photo alone is working with part of the evidence.
What this means for you: read the label as a sentence about the mouth, not as a claim about the person. The readings and the label answer different questions, and only one of them is in the table.
What the readings change for liner, gloss and photographs
Liner is the place where the bow becomes visible. The two peaks and the central dip are fixed points, and a line drawn along the wet border reproduces the measured outline. A line drawn above the peaks makes the bow look taller than the reading; one drawn across the dip fills it in. On a mouth the guides would call downturned, extending the line along the existing corner angle keeps that angle visible, while lifting the end changes what a photograph shows. None of those choices changes the measurement; they change what a viewer sees first.
Gloss works on the apparent height of the vermilion. A highlight sitting on the lower lip makes that lip read taller in a photograph, and a highlight on the upper lip does the same for the upper. If you want the four readings to match the photo, take the measurement photo before the gloss. If you want the photo to show a different balance, the gloss is the tool, and the reading underneath stays the same.
Photographs add a distance problem. The Indonesian study used a one-metre camera distance and a ruler in the frame. A close phone selfie exaggerates the mouth relative to the rest of the face, and a slightly tilted head changes the apparent corner angle. For a before-and-after comparison, keep the same distance, the same light and the same neutral expression, and put a ruler in the frame so the scale is not guessed.
What this means for you: the readings describe the mouth at rest; liner, gloss and camera distance describe how it is seen. Keep the two descriptions separate and neither one has to pretend to be the other.
A lip shape is a vocabulary, not a measurement. The vocabulary is useful when it names its axis: a wide mouth, a bottom-heavy balance, a defined bow, a downturned corner. It is less useful when it is asked to carry a verdict. The four readings are the part a ruler can settle; the rest is a description you can check.
Start with the readings, write the method beside them, and use the label only for the axis it names. A number and a name answer different questions; this page keeps them apart on purpose.
Questions this page answers
What are the main lip shapes, and why do lists disagree?
The two 2026 guides read for this page both use nine labels: full, thin, wide, round, heart-shaped, bow-shaped, top-heavy, bottom-heavy and downturned. They look like nine alternatives, but they answer different questions: thickness, balance, width, bow, outline and corner direction. One mouth can carry several of those labels at once, which is why a single list can feel incomplete.
How can I measure my lip shape with a ruler?
Take four distances at rest: mouth width from corner to corner, upper vermilion height from the lowest point of the upper wet border to the lip line, lower vermilion height from the lip line to the midpoint of the lower wet border, and cupid's-bow height from the central dip to one peak. Repeat each reading on two days, keep the average, and write down where the ruler sat.
Is there a preferred upper-to-lower lip ratio?
A 2017 preference study had 428 raters rank synthetic faces and choose a 1:2 upper-to-lower ratio, which is 0.5. The measured means in the studies on this page run from 0.56 to 0.84, so none of those samples sits at 0.5. A rater preference and a population mean are different findings, and neither is a rule for an individual mouth.
What counts as a cupid's bow, and is it measured?
A cupid's bow is the double curve of the upper vermilion border, with two peaks called crista philtri and a central dip at labiale superius. Researchers measure it as a width between the peaks and a height from the dip to a peak. One study of 400 models reported a mean width of 12.3 mm and a mean height of 8.2 mm. One 2026 guide calls a bow pronounced above 3.5 percent of mouth width, about 1.7 mm on a 49 mm mouth, but that is its own rule; the measurement studies give no such cutoff.
Does lip shape change how lipstick or gloss looks?
It changes where a line or a highlight sits, not the measurement underneath. Liner drawn along the wet border follows the measured outline, while a line above the peaks changes the apparent bow height. Gloss on the lower lip makes that lip read taller in a photograph. Both are ways of changing what a viewer sees first; the four readings describe the mouth at rest.
Can a frontal photo tell me my lip shape?
A frontal photo can show mouth width and the two vermilion heights if there is a ruler in the frame for scale, and it can suggest the bow and the corner angle. It cannot measure how far the lips project in profile, and a close selfie changes the proportions. The Indonesian study cited here used a one-metre distance and a scale in the frame, so a photo taken the same way is more comparable than a close-up.