The science behind diagonal reading: what eye-tracking research actually shows

The science of how humans read has advanced considerably since the early twentieth century. What we now know — from millisecond-precision eye-tracking studies, neuroimaging research, and computational modelling — gives us a much clearer picture of why diagonal reading works, why it has limits, and what "skilled scanning" actually looks like at the level of eye movements.

This post summarises the key evidence, with citations.

Eye movement basics

Reading is not a smooth, continuous glide of the eyes across text. It is a series of rapid jumps — saccades — alternating with brief pauses — fixations — during which the visual system actually acquires information.

Rayner (1998) established the core statistics: a typical skilled adult reader makes fixations averaging 200–250 ms each, with saccades spanning 7–9 letter positions on average. Shorter fixations and longer saccades = faster reading. But longer saccades also mean more words are skipped, which only works if the reader can predict or infer what was skipped from context.

Crucially, the perceptual span during a fixation — the region from which useful information is extracted — extends about 14–15 character positions to the right of the current fixation point and only 3–4 positions to the left (in left-to-right reading languages). This asymmetric window is why skilled readers can process upcoming words in parafoveal preview: you partially process the next word before your eye lands on it. RSVP reading disrupts this; diagonal reading preserves it.

Rayner et al. (2016): the landmark review

The most comprehensive review of reading speed research is Rayner, Schotter, Masson, Potter, and Treiman (2016), "So Much to Read, So Little Time: How Do We Read, and Can Speed Reading Help?" published in Psychological Science in the Public Interest.

Key findings relevant to diagonal reading:

1. Skipping is normal, but not unrestricted. Skilled readers skip about 30–35% of words during normal reading — mainly short function words. This is automatic and below conscious awareness. Skipping content words, by contrast, reliably reduces comprehension.

2. Speed and comprehension trade off. There is no technique that allows readers to substantially increase speed without some comprehension cost. However, the size of the cost depends on content familiarity: for highly familiar topics, skilled readers can skim at 500–700 WPM with moderate comprehension loss; for unfamiliar dense material, the loss is severe even at 400 WPM.

3. Regressions serve a purpose. Regressions (moving the eye back to earlier text) are not a bad habit — they occur when comprehension fails and repair is needed. Suppressing regressions without addressing the underlying cause simply leaves comprehension gaps unnoticed.

4. RSVP disrupts parafoveal preview. Because RSVP shows one word at a time at a fixed point, it eliminates the reader's ability to preview upcoming words. This is the mechanism by which RSVP hurts comprehension at high speeds. Diagonal reading, by contrast, preserves the normal left-to-right scan and thus preserves parafoveal preview.

Nielsen (2006) and the F-pattern

Jakob Nielsen's 2006 eyetracking study — conducted at the Nielsen Norman Group — remains one of the most cited pieces of evidence for how readers scan digital and printed text in practice.

The study tracked 232 participants reading websites. The dominant pattern that emerged was an F-shape:

• A horizontal movement across the top of the content area (the top bar of the F)
• A shorter horizontal movement further down the page, usually covering a shorter area than the first (the lower bar)
• A vertical movement down the left side of the content (the spine of the F)

The practical implication: the first sentence of each paragraph, and the first word of each line, receives the most fixation time. The right half of paragraphs beyond the first line receives very few fixations in this pattern.

Nielsen and Pernice (2010) extended this in Eyetracking Web Usability, identifying additional patterns (layer cake, spotted, bypass, commitment) and noting that the F-pattern is especially common for long prose where the reader's goal is to find something specific rather than read carefully.

Critically, Nielsen's research describes natural scanning behaviour — readers defaulting to F-pattern when untrained. Diagonal reading is a deliberate refinement of this natural tendency: rather than letting the vertical component of the F drift randomly, diagonal reading gives it a structured, reproducible path that ensures more even coverage across the full text.

Zipf's law and content word density

George Kingsley Zipf's 1949 Human Behavior and the Principle of Least Effort introduced what is now called Zipf's Law: in any natural language corpus, word frequency follows a power law distribution where the most common word appears approximately twice as often as the second most common, three times as often as the third, and so on.

The practical implication for reading: the most frequent words in English (the, a, of, in, is, was, to, that...) account for a disproportionate share of total word tokens in any document — yet carry very little semantic content. Function words (articles, prepositions, conjunctions, auxiliaries) are predictable from context. Content words (nouns, main verbs, adjectives) carry the information.

This is the linguistic basis for diagonal reading: if you reliably fixate on content words and your brain automatically fills in the grammatical scaffolding from context, you get the semantic payload of the text with a fraction of the fixations.

Adler's (1972) How to Read a Book — pre-dating the eye-tracking literature but consistent with it — made the same point from a pedagogical angle: skilled inspectional reading involves targeting the "skeleton" of the text (structure, main claims, key evidence) rather than the "flesh" (elaboration, examples, transitions). The skeleton is, approximately, the content word layer.

Just & Carpenter (1987): individual differences in eye movements

Just and Carpenter's The Psychology of Reading and Language Comprehension (1987) documented systematic individual differences in eye movement patterns between skilled and less-skilled readers. Skilled readers show:

• Fewer regressions (5–10% vs 15–20% for less-skilled readers)
• Shorter fixation durations on common words
• Larger saccade lengths — skipping more predictable words
• Better use of parafoveal preview — partially processing words before the eye arrives

These differences are not primarily innate — they reflect learned strategies developed through extensive reading practice. This is the empirical basis for the claim that diagonal reading can be trained: the eye movement patterns of skilled scanners are learnable, not fixed.

Kintsch (1988): what comprehension actually requires

Walter Kintsch's construction-integration model (1988) describes reading comprehension as a two-stage process: constructing a network of propositions from the text's surface features, then integrating those propositions with prior knowledge to build a coherent mental model (the situation model).

The implication for diagonal reading: the situation model can be partially constructed from a subset of propositions if the reader has sufficient prior knowledge to fill in the gaps. For a reader who already knows the topic well, diagonal reading samples enough propositions to let prior knowledge complete the situation model. For a reader with little prior knowledge, there are too many gaps — comprehension fails.

This explains why expert readers in a domain can skim academic papers in that domain far more effectively than a non-expert skimming the same paper. It also explains the consistent finding in the research that the benefit of skimming strategies is highly domain-specific.

Summary for practitioners

The evidence supports the following practical conclusions:

1. Diagonal scanning works because language is redundant (Zipf), skilled readers already skip 30–35% of words naturally (Rayner et al.), and readers naturally default to F-pattern scanning of long documents (Nielsen).

2. It has real limits: comprehension drops significantly when content is unfamiliar (Kintsch), regressions serve a repair function that skimming can't replicate (Rayner et al.), and the technique cannot achieve the extreme speeds claimed by commercial speed reading programmes.

3. It can be trained: eye movement efficiency improves with practice (Just & Carpenter), and the difference between skilled and novice scanners is partly attributable to learned strategies rather than innate ability.

4. The right frame: treat diagonal reading as a first pass strategy for triage and extraction, not a replacement for careful reading of complex or unfamiliar material.

The WarpRead diagonal reader tool visualises the diagonal scan path across any pasted text, letting you see exactly which words your eyes are targeting at different density and step settings. The diagonal reading course covers all five of the evidence-based skills described in this post.

References

• Adler, M. J., & Van Doren, C. (1972). How to Read a Book (revised ed.). Simon & Schuster.
• Just, M. A., & Carpenter, P. A. (1987). The Psychology of Reading and Language Comprehension. Allyn & Bacon.
• Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-integration model. Psychological Review, 95(2), 163–182.
• Nielsen, J. (2006). F-Shaped Pattern for Reading Web Content. Nielsen Norman Group.
• Nielsen, J., & Pernice, K. (2010). Eyetracking Web Usability. New Riders.
• Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124(3), 372–422.
• Rayner, K., Schotter, E. R., Masson, M. E. J., Potter, M. C., & Treiman, R. (2016). So much to read, so little time: How do we read, and can speed reading help? Psychological Science in the Public Interest, 17(1), 4–34.
• Zipf, G. K. (1949). Human Behavior and the Principle of Least Effort. Addison-Wesley.