Skip to content
Back to search results

Eye-Tracking in Reading

Introduced by
Keith Rayner and the eye-movement reading-research tradition
Year
1998
Type
Online measure of reading processing from eye movements
Family
Psycholinguistics of reading
Font size

What does it do?

Eye-tracking in reading records where readers look and for how long while they read text naturally, turning the eyes into a continuous index of comprehension.

Reading is not a smooth glide but a sequence of brief fixations punctuated by rapid saccades and occasional regressions back to earlier words. By logging this pattern with millisecond precision, researchers derive measures — first-fixation duration, gaze duration, total reading time, regression-path duration, skipping rate — that reveal, region by region and stage by stage, where and how much the language system struggles. Established by Keith Rayner's research program, it is the gold-standard online measure of natural reading.

The intuition behind the method

Your eyes go where your mind is working. Reading feels continuous, but the eyes actually hop from word to word, pausing longest on words that are hard to identify or integrate, skipping ones that are short and predictable, and darting backward when something does not add up. Each of those behaviors is a clue: a long first fixation says a word was hard to recognize, a regression says comprehension broke down and the reader is searching for the fix. Eye-tracking reads these clues off natural reading without asking the reader to do anything unusual, giving a moment-by-moment, fine-grained record of the mind at work behind the page.

History

Eye movements in reading have been studied since the late nineteenth century, but modern psycholinguistic eye-tracking is defined by Keith Rayner's decades-long program. His 1998 Psychological Bulletin review synthesized twenty years of research, codifying the fixation/saccade/regression vocabulary and the battery of reading measures (first-fixation, gaze, go-past, total time, skipping) that remain standard, and establishing how eye movements index moment-to-moment language processing. His 2009 review extended the account across reading, scene perception, and visual search. The paradigm is distinct from broader eye-tracking-analysis and from psycholinguistic eye-tracking in the visual-world (spoken-language) sense: it specifically concerns the eyes during natural reading of text, and it remains the benchmark online reading measure against which cheaper methods are validated.

When to use it

Use eye-tracking in reading when you need a fine-grained, naturalistic, online picture of reading that distinguishes early lexical processing from later integration and recovery — for studying lexical access, syntactic and semantic integration, garden-path reanalysis, predictability, and reading development or disorders. It is the method of choice when naturalness and stage decomposition matter and when regressions and rereading are part of the phenomenon. It is less appropriate when budget, hardware, or large remote samples rule out an eye-tracker (where self-paced reading is the practical alternative), when the question is about production rather than reading, or when participants cannot tolerate the setup.

Limitations and considerations

  • Requires expensive, calibrated equipment and trained operators, and most high-precision systems are lab-bound.
  • Eye movements are driven by low-level factors (word length, launch position, predictability) that must be carefully controlled.
  • The many available measures invite analytic flexibility and selective reporting if not pre-specified.
  • Mapping eye-movement measures onto specific cognitive stages is theory-dependent and not always clear-cut.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Records natural, silent reading without an artificial task, preserving normal eye-movement behavior.
  • High temporal and spatial resolution captures fixations, saccades, regressions, and skipping in fine detail.
  • Yields multiple measures that dissociate early (lexical) from late (integrative) processing on the same region.
  • Captures rereading and regressions — recovery from comprehension breakdown — that serial reading tasks cannot see.

Applications

  • Studying lexical access through word-frequency, length, and predictability effects on fixation durations and skipping.
  • Investigating syntactic processing and garden-path reanalysis via regressions and go-past times at disambiguation.
  • Testing semantic and discourse integration effects in sentence and passage reading.
  • Examining reading development, dyslexia, and aging by comparing eye-movement profiles across populations.
  • Validating predictability and surprisal estimates from language models against natural-reading fixation patterns.

How it works

How it works

1

Sentences or passages are written so that a target word or region varies across conditions while the surrounding text is matched on length, frequency, and predictability. The text is laid out for display, and interest areas (AOIs) are defined over the critical region and relevant pre- and post-target regions. Critical words are kept away from line boundaries, and word length, frequency, and cloze predictability — all strong drivers of eye behavior — are controlled or measured.

2

A calibrated eye-tracker samples gaze position at high frequency while the participant reads silently and naturally, with a chin/forehead rest for high-precision systems. The raw gaze stream is parsed into fixations (eyes relatively still, taking in information) and saccades (rapid moves), and progressive versus regressive movements are distinguished. Some regions are not fixated at all, so the per-region skipping rate, \(\text{skip rate} = \#\{\text{trials region not fixated}\}/\#\text{trials}\), is itself a measure, higher for short, predictable words.

Formula
\text{skip rate} = \frac{\#\{\text{trials region not fixated}\}}{\#\text{trials}}
3

From the fixation sequence over each interest area, standard measures are computed. Early measures index lexical access — first-fixation duration (the first fixation on the region) and gaze duration, the sum of all first-pass fixations before leaving the region, \(\text{gaze duration} = \sum_i f_i\). Later measures index integration and recovery — go-past (regression-path) duration, the time from first entering a region until moving past it including any regressions, and total reading time, all fixations on the region. The split between early and late measures is what lets reading eye-tracking separate stages of processing.

Formula
\text{gaze duration} = \sum_i f_i \;\text{(first-pass fixations on region)}
4

Each measure is analyzed by condition with linear (or, for binary skipping, logistic) mixed-effects models, with participants and items as crossed random effects and word-length/frequency/predictability as covariates. The profile across measures is interpreted: an effect confined to first-fixation and gaze durations implicates early lexical processing, whereas an effect emerging only in go-past and total times implicates later integration or reanalysis — Rayner's framework for reading from the eye-movement record.

Common pitfalls

  • Placing the critical word near a line boundary, where return sweeps and wrap-up distort fixation measures.
  • Failing to control or covary word length, frequency, and predictability, which dominate fixation durations and skipping.
  • Cherry-picking among first-fixation, gaze, go-past, and total-time measures after seeing the data, inflating false positives.
  • Treating an eye-tracking study like self-paced reading and reporting one undifferentiated 'reading time' instead of the staged measures.

Frequently asked questions

How is eye-tracking in reading different from eye-tracking analysis or visual-world (psycholinguistic) eye-tracking generally?

All record gaze, but the object and measures differ. The general 'eye-tracking-analysis' method covers gaze recording and analysis broadly (UX, scene viewing, marketing), and 'psycholinguistic-eye-tracking' in its visual-world form tracks where listeners look at objects in a scene while hearing spoken language. Eye-tracking in reading specifically concerns the eyes moving over written text, and it uses reading-specific measures — first-fixation duration, gaze duration, go-past time, total time, regression and skipping rates — derived from fixations on words. This page is the reading paradigm; consult those other methods for scene/visual-world designs.

What is the difference between early and late eye-movement measures?

Early measures — first-fixation duration and gaze duration (first-pass time on a region) — reflect initial word identification and lexical access. Late measures — go-past (regression-path) duration and total reading time — reflect later processes such as syntactic and semantic integration, ambiguity resolution, and recovery from misparsing, because they include rereading and regressions back to the region. Locating an effect among these measures tells you which stage of processing a manipulation affects, which is a key advantage of reading eye-tracking over coarser online methods.

Why do readers skip some words and regress to others?

Skipping occurs mostly for short, high-frequency, and predictable words that can be identified parafoveally before the eyes land, so skipping rate is itself a sensitive online measure of predictability and ease. Regressions — backward saccades to earlier text — occur when comprehension breaks down, as at a garden-path disambiguation or an implausible continuation, and the reader returns to recover meaning. Both behaviors are informative: higher skipping signals easy processing, while regressions and longer go-past times signal difficulty and reanalysis.

Related methods

Sources

  1. Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124(3), 372–422.

    Open DOI reference
    Copied.
  2. Rayner, K. (2009). Eye movements and attention in reading, scene perception, and visual search. Quarterly Journal of Experimental Psychology, 62(8), 1457–1506.

    Open DOI reference
    Copied.