FWHR test: facial width-to-height ratio

1. 1

   Take one straight-on photo

   Face the camera with the lens at eye level and keep a neutral expression. Step back from a close front camera, because a near lens widens the middle of the face.

2. 2

   Let the browser measure it

   Choose the photo on this page. The landmark model loads on your device and divides the visible cheekbone width by the upper-face height; the file is not uploaded or logged. Only explicit consent creates a local 30-minute checkout draft.

3. 3

   Read the ratio as a description

   The number describes proportions in this photograph. Compare it only with a value measured the same way, and read no personality claim into it.

A common approach divides bizygomatic width—the visible span near the cheekbones—by an upper-face height. Studies and tools do not always choose the same vertical endpoints, so a value from one method should not be compared blindly with another.

FWHR has appeared in research about perceived dominance and threat, but the evidence for behavioral claims is mixed. A meta-analysis found small sex differences and relationships with some perceptions while describing mixed support for broader claims.[1]

Do not use a face ratio to infer aggression, trustworthiness, leadership or moral character. Those leaps turn a descriptive proportion into physiognomy.

• A close lens enlarges central features relative to the edges.
• Yaw reduces the visible width.
• Hair and soft tissue affect the apparent cheek boundary.
• Expression can shift the brow and upper lip landmarks.
• Body composition can affect visible width without changing the skull.

The width in FWHR is often the visible bizygomatic span, but images do not reveal the underlying bony landmarks as cleanly as a three-dimensional scan. The height may run from the brow to upper lip, brow to mouth or another upper-face boundary depending on the paper or tool. Small endpoint changes matter because the measured vertical distance is relatively short.

A reported value is incomplete without its construction. When reading research, check the landmark definitions, whether raters placed points manually, how head pose was standardized and which population was sampled. When using a calculator, compare only outputs from the same method. There is no universal “ideal FWHR” independent of those choices.

• Confirm the width endpoints.
• Confirm the vertical endpoints.
• Check whether hair or ears affect the crop.
• Look for pose standardization.
• Avoid transferring a threshold between methods.

Perspective distortion is strongest when a camera is close. Central features sit nearer the lens while the sides of the face fall farther away, changing their relative apparent sizes. Moving back and using modest optical zoom generally produces a flatter-looking representation. Head rotation also shortens the visible width even when the underlying face is unchanged.

Screenshots, profile-leaning selfies and heavily cropped images are poor inputs for proportion tracking. Image resolution alone does not solve perspective. A high-resolution close selfie can be less representative than a modest-resolution portrait taken farther away with the face square to the camera.

• Step back from the front camera.
• Keep both sides similarly visible.
• Use the same lens for comparisons.
• Avoid stretching or aspect-ratio edits.
• Keep the full cheek width in frame.

Group-level correlations, when they appear, do not support judging an individual. Effects can be small, sensitive to study design or tangled with sex, body size, age, expression and observers’ stereotypes. A face ratio cannot reveal whether one person is aggressive, trustworthy, generous or competent. Treating it as a character detector repeats the errors of physiognomy.

For an individual, FWHR can only describe how the chosen visible width compares with the chosen height in a photograph. Omogle makes no behavioral prediction from that ratio and sends no face-derived category to analytics.

First identify whether the study measures actual faces or observers’ impressions, because perceived dominance and measured behavior are not the same outcome. Check sample size, age, sex composition, ancestry, image standardization and whether the analysis was planned before the data were seen. A result in one group may not generalize to another.

Next, separate an average association from a prediction about one person. Two groups can differ on average and still overlap so much that the ratio tells you nothing useful about an individual. Check replication, effect size and uncertainty rather than relying on the headline.

Finally, notice whether face size, body size or other correlated variables were considered. A ratio can inherit relationships from its numerator, denominator or sample selection. That complexity is another reason Omogle reports FWHR as a visible proportion and refuses to turn it into a behavioral profile.

If a calculator gives an “ideal” value, check where it came from. A target chosen for one landmark convention or sample may not transfer to another. The useful comparison is between photos measured by the same method.