Facial Thirds: Measure Yours and See What Real Faces Look Like

The rule is about two thousand years old. The Roman architect Vitruvius, writing about proportion in temples and in the human body, described the face as three equal parts, and Renaissance artists such as Leonardo da Vinci ruled lines across their face studies in the same spirit. It was a tool for drawing a convincing head, not a description of measured people.

Modern anthropometry inherited the rule as one of the “neoclassical canons” and then tested it. The short version of what it found: real faces vary a lot, the middle third is usually the shortest, and the canon is a poor yardstick for anyone.

When orthodontic textbooks and looksmaxxing forums talk about facial thirds today, they usually mean the same three bands Vitruvius described, measured between slightly different points. That small difference in landmarks turns out to matter, as section 5 explains.

Switch on the Thirds layer and you'll see up to four horizontal lines on your photo, labelled hairline, brows, nose base and chin.

The brow line is taken at the point between your eyebrows on the midline, not at the top of the arches. The nose base is the point where the nose meets the upper lip. The chin line is the lowest point of the chin on the face outline. All three come from Google's MediaPipe face landmarks, after the image has been rotated so your pupils sit level and your head has been turned back to frontal.

The hairline is the hard one, because there's no landmark for it. We run a second Google model, a hair-segmentation model that marks which pixels are hair. Starting at the top of the face mesh, the scan walks straight up the midline and stops at the first row where most of a small strip of pixels is hair. If it never finds one within a sensible distance, it reports the hairline as not found and skips every measurement that needs it, rather than guessing.

Your upper, middle and lower thirds are then shown as percentages of the hairline-to-chin length, each with a ± range from the photo plus three re-framed copies.

If your hair covers your forehead, the segmentation model sees hair where your forehead is, and the line it finds is the edge of your fringe. Pull bangs back with a clip or headband for the measurement photo. Our rulebook lists this as a stylist convention for every face scan: hair over the hairline or jaw hides the outline the measurements depend on.

If your hairline is receding or very high, the scan measures where hair starts in your photo today. It cannot know where your hairline used to be, so a high upper third may reflect hair, not bone. That's worth knowing before you read anything into the number.

If no hairline is found at all, you still get one useful proportion: middle ÷ lower third. It compares brows-to-nose-base with nose-base-to-chin, so it needs no hairline. 1.0 means the two are equal; below 1.0 means the lower third is longer.

Researchers have tested the thirds canon directly, by measuring faces with calipers or 3D cameras and counting how many fit within a small tolerance. Here is what they found.

If thirds aren't equal, how are they arranged? The measurement data shows a consistent pattern: the middle third is usually the shortest. Farkas and colleagues, as reported by Al-Sebaei (2015), found the lower third longer than the middle third in 100% of North American white and African-American adults they measured, and the upper third longer than the middle in 95% of North American whites and 100% of African Americans.

The average measurements tell the same story. In Farkas's North American white sample, as reproduced by Virdi and colleagues (2019), hairline to nasal root averaged 67.1 mm in men and 63.0 mm in women, nasal root to nose base 54.8 and 50.6 mm, and nose base to chin 72.6 and 64.3 mm. In Al-Sebaei's Saudi sample the three thirds averaged 64.71, 54.12 and 65.02 mm in men and 65.46, 53.19 and 60.27 mm in women.

Notice also how much the upper third varies. In the Farkas data its standard deviation (the typical spread) is 7.5 mm in men and 6.0 mm in women, about twice the spread of the middle third (3.3 and 3.1 mm). Part of that is simply where hair starts, which is one reason a “big forehead” is such a common worry and such a poor measure of anything.

For the lower third alone, a 2025 photogrammetry study of 400 Caucasian female professional models found it averaged 0.32 ± 0.02 of total facial height (Winiarska and colleagues). That is a sample of models, not the general population, and the authors' landmarks differ from ours.

Even a perfectly measured photo won't reproduce those millimetre figures, for three reasons. First, the landmarks differ: the studies above start the middle third at the nasal root (the deepest point between the eyes), while our brow line sits at the point between the eyebrows, a little higher. Second, they used calipers or 3D cameras on the real face, while a phone photo is a 2D projection; close range enlarges the centre of the face, and Pressler and colleagues (2022) measured nasal length 6.4% longer in 12-inch selfies than in clinical photos. Third, the landmark model has its own error, about 2% to 3.3% of the distance between the eyes on average, per Google's model card.

That's why FaceCheckup shows your thirds as plain numbers, without a “normal range” beside them. The published norms use different landmark definitions, so telling you that your 36% upper third is normal or abnormal against them would be making something up. Compare your own photos with each other instead: if two good photos agree within their ± ranges, that's your measurement.

Vertical proportions are the most sensitive to how you hold your head. Tipping your chin up or down foreshortens parts of the face in a flat photo and moves every line. The quality check warns at 10° of chin tilt and asks for a retake beyond 18°.

Get these steps right and your facial thirds should repeat closely from one photo to the next; get them wrong and the lines can move by more than the differences you're trying to see.

This ratio is the most robust part of the thirds measurement, because it doesn't depend on hair. A value of 1.00 means your brows-to-nose-base and nose-base-to-chin distances are equal. A value of 0.85 means the middle third is 85% as long as the lower third.

In the published averages above, the middle third is shorter than the lower one. Dividing the published means ourselves gives about 0.75 for Farkas's North American white men and 0.79 for women, and about 0.83 and 0.88 for Al-Sebaei's Saudi men and women. Because those studies measure from the nasal root, not from between the brows, those figures describe the populations, not a target for your scan. No study we found identifies an attractive or ideal value for this ratio.

Expression matters here too. A smile pulls the mouth corners up and back and can change the lower third by more than the ± range, which is why the quality check flags smiling and an open mouth.

If you want to track it over time, for example while trying a new beard shape or fringe for photos, keep the camera in the same place and note the ± range each time. A change smaller than the range is noise, not a change in your face.

Proportions aren't problems, but they do guide styling choices, and that's where measuring is genuinely useful. Every suggestion below comes from our sourced rulebook, with its label.