Eye tracking accuracy: how it is measured

Accuracy is the single most important parameter in any eye tracking study. If the system does not record where participants actually look, everything else rests on unstable ground. Because every setup is different — from lab monitors to eye-tracking glasses in motion — there are several ways to determine a system’s accuracy. This article reviews the most common methods: each has strengths and limits, and understanding those trade-offs helps you pick the right approach for your research.

Pixel error

Pixel error is one of the most common ways to measure accuracy. It indicates how good the calibration was: it is quantified as the average (or maximum) number of pixels by which the measured gaze deviates from the actual calibration points. It is the difference between the real calibration point — a dot on the screen, say — and where the system believes the person was looking.

• A lower pixel error means a more accurate calibration.
• For example, if the calibration point is at (500, 400) and the system detects gaze at (505, 398), the error would be around 5 or 6 pixels.

Reporting accuracy in pixels can produce misleadingly high values on small screens, even when the data is good. This comes down to the different pixel density (PPI) between desktop monitors and phones or tablets: the smaller the device, the higher the PPI. For reference, a modern phone packs very high resolution into a small screen, while a desktop monitor spreads its pixels across a much larger surface. In other words, phones and tablets tend to show a larger pixel error than a monitor, simply because they pack the same resolution into a smaller frame.

Why measure in pixels?

For screen-based eye tracking systems, reporting accuracy in pixels is usually the most practical option. Since the participant’s gaze is always directed at a defined screen, pixel error directly reflects how far the measured gaze point is from the target on that same screen. That makes interpretation straightforward: if the error is 20 pixels, you immediately know how far the estimate strays within the digital content — a website, a video or an interface. Because areas of interest (AOIs) are also defined in pixels, the error scales naturally with them and gives a direct sense of how it affects AOI-based analysis.

Accuracy in multi-camera systems

If you work with a system that builds a model of the environment using several cameras, you are no longer limited to a flat screen as the reference. In those setups, gaze accuracy is defined relative to objects and distances in physical space: instead of pixels, deviations are expressed directly in real units such as centimeters or meters.

Angular error

Angular error is another widely used measure. It describes the angular distance, in degrees of visual angle, between where the person was actually looking (the calibration target) and where the system estimated their gaze to be.

• During calibration, the participant looks at specific points on the screen or in the environment.
• The system estimates the gaze position.
• Angular error is the angular distance between the real point (where they should have looked) and the measured point (where the system believes they looked).

Why measure in degrees rather than pixels?

If accuracy were reported only in pixels, the result would be tied to one specific resolution, screen size and viewing distance. The same error could look very different depending on whether the person sits close to a small laptop or far from a large monitor. Expressing accuracy in degrees of visual angle makes the measure independent of the setup. An angular error of 1° represents the same deviation for the participant’s eye, whatever the screen. In practice, 1° equals roughly 1 cm at 57 cm of distance, or about 2 cm at 114 cm. That makes angular error a consistent standard for comparing accuracy across studies. Typical values:

• High-quality desktop eye trackers: around 0.3° to 0.5° of angular error.
• Mobile or glasses-based systems: often 0.5° to 1.0°, or more.

What can affect eye tracking accuracy

It is worth understanding what "accuracy" really means. As mentioned, on small screens it can look like accuracy drops, but that is only an effect of how pixel deviation appears on small displays: it does not mean the system is less accurate. That said, there are real-world factors that do affect how well an eye tracker measures gaze. The most common ones:

• Glasses with strong prescriptions or progressive lenses: they distort light and make it harder to read gaze correctly.
• Glasses with infrared-blocking coatings: since most eye trackers use infrared light, these lenses reduce accuracy or prevent recording altogether.
• Dirty or highly reflective glasses: smudges, grease or strong glare interfere with the sensors.
• Eye conditions: people with nystagmus or similar conditions may calibrate poorly.
• Things covering the face: wearing a cap and a face mask at the same time can obscure the eyes and affect recording.

Best practices for researchers

To get the most accurate results, a few practical steps help:

• On small screens: check the calibration manually before starting, asking the participant to fixate on specific points.
• Participants with glasses: exclude anyone wearing infrared-blocking or progressive lenses; the rest can be included or excluded depending on calibration quality.
• Participants with certain conditions: those who have had eye surgery or have nystagmus usually calibrate poorly and are generally best excluded.
• Ambient lighting: use controlled light that avoids infrared sources. Overhead or frontal light works best, but avoid glare on glasses.
• Ergonomics: use a fixed chair to minimize head movement, combined with a height-adjustable desk that adapts to each participant.