Are Learning Styles Real? What the Evidence Actually Says

TL;DR

• The VARK questionnaire was built from its authors’ own observations, and they wrote in the original 1992 paper that testing it for validity and reliability was “unnecessary and inappropriate”.
• Roughly 89% of educators across 18 countries believe that matching teaching to a student’s style improves learning. The evidence for that specific claim has never arrived.
• Most debunks overstate the case. Pashler’s team did not find zero supporting studies. They found one, and explained why it was not convincing.
• A 2024 meta-analysis did find a small positive effect for matching. We read 14 ranking pages on this question and none of them mentioned it. We do, and we explain why it still does not justify the practice.
• Preferences are real. The instruction that follows from them is not the useful part.

Where the four letters came from

VARK — visual, aural, read/write, kinesthetic — was published in 1992 by Neil Fleming and Colleen Mills. Fleming had spent nine years as a senior schools inspector in New Zealand before moving into faculty development, and the questionnaire came out of what he had seen.

That is not an unkind reading. It is what the paper says. The questions were “drawn from observations students had shared, personal observations of our own modal preferences, and preferences reported by friends”. And on the matter of validation, the authors are unusually direct:

Read in context, that is a reasonable position. Fleming and Mills built a conversation starter for faculty development and said so. What happened afterwards — the inventory becoming a diagnostic tool, then a basis for lesson planning, then a commercial product line — was not their design.

Thirty-four years later, a systematic review of 37 studies across 18 countries found that 89.1% of the 15,405 educators surveyed agreed that matching instruction to a learner’s style improves outcomes. The review’s own authors caution that most of those surveys used convenience samples, and that it is not always clear respondents understood they were being asked about matching rather than about preferences existing at all.

What “matching works” would have to look like

In 2008, Pashler, McDaniel, Rohrer and Bjork set the bar that the field still uses. To show that matching helps, a study has to do four things: sort people into style groups, randomly assign people within each group to different teaching methods, test everyone with the same assessment, and then find a crossover — group A does better with method A, and group B does better with method B.

That last word is where the whole question lives. If both groups do better with diagrams, diagrams are simply better. That is a finding about the material, not about the learners.

Here is the part almost every article gets wrong. Pashler’s team did not report finding nothing. They reported finding exactly one study that “could be described as even potentially meeting” the criteria — Sternberg and colleagues, 1999 — and that study was positive. They then explained why they did not find it convincing: the interaction appeared only in derived composite measures, raw scores by condition were never published, only about a third of the sample could be classified into a dominant style at all, and the effect emerged after outliers were dropped for reasons the paper does not give.

“We found one and here is why it does not hold” is a stronger claim than “we found none”, because it survives someone reading the original.

The debunks are out of date too

On 27 July 2026 we searched four queries — are learning styles real, are learning styles a myth, are learning styles debunked, learning styles evidence based — and opened every unique result. Three could not be read (a firewall, a paywall, a rate limit) and were dropped rather than guessed at. Of the 14 pages we could read:

This matters more than a scoring exercise. A reader who has been told “there is not one single study” and then encounters a 2024 meta-analysis reporting a significant positive effect has been handed a reason to distrust the entire debunk. Overstating a strong case is how you lose it.

The meta-analysis that found an effect

In 2024, Clinton-Lisell and Litzinger pooled 21 studies, 101 effect sizes and 1,712 participants, and reported a significant benefit for matching: g = 0.31, 95% CI [0.05, 0.57], p = .02. No sign of publication bias. This is the first meta-analysis of the matching hypothesis to come out positive, and any honest treatment of the question has to deal with it rather than skip it.

Four things in the paper keep it from rescuing the practice, and three of them are the authors’ own points.

• The crossover pattern showed up in 11 of 42 outcome measures. The rest of the positive average comes from studies where one method was better for everyone — a modality effect, not a matching effect.
• The studies producing crossovers are tiny. Eighteen participants. Twenty-two. Thirty-seven. Only 5 of the 21 studies met What Works Clearinghouse standards.
• 85% of the studies excluded participants whose questionnaire scores did not place them clearly in one style. The effect is measured on an artificially polarized remainder.
• The size is unimpressive next to the alternatives. The authors compare it themselves: removing extraneous material from a lesson gives g = 0.33, segmenting it gives g = 0.32, and the plain modality effect gives g = 0.70. Their own conclusion is that the benefits are “too small and too infrequent to warrant widespread adoption”.

Running alongside it, a 2025 analysis by Hattie and O’Leary sorted 17 meta-analyses covering roughly 700 studies and over 100,000 learners into two piles. The four that actually tested matching — 143 studies, 401 effect sizes — averaged d = 0.04. The thirteen correlational ones averaged r = 0.24, and it is those that keep the overall picture looking respectable. Their explanation for the myth’s survival is that correlational work repeatedly confuses a style with a strategy: what a learner does is measured, and then labelled as what a learner is.

One more result, reported inside the 2024 meta-analysis, is worth the price of admission. When researchers gave students a style questionnaire and then told them a randomly assigned result, students did better when the material matched the fabricated style. Matching their real questionnaire result did nothing. Whatever the benefit is, it travels with expectation.

Three claims that get merged into one

Most of the confusion in this argument comes from treating one question as three. They have very different amounts of evidence behind them.

The same pattern shows up outside the lab. Following 426 anatomy students through a semester, researchers found that most did not even study in the way their VARK profile suggested — and the ones who did got no better grades for it.

What to do instead

The shift: stop asking what kind of learner you are and start asking what the material demands. The question “am I a visual learner?” has no useful answer. The question “does this content need a diagram?” has one, and it changes from topic to topic.

The first move: take whatever you are studying this week and, before you open it, write down how you will test yourself on it. Not review it — test. That single change has more evidence behind it than every style inventory ever printed.

The techniques that survive scrutiny work for everyone, which is exactly why they are less appealing than a personality result:

• Retrieval practice. Closing the book and reconstructing the material beats rereading it, and it beats how rereading feels by an even wider margin. The catch: it is unpleasant, and people abandon it because it feels like failing.
• Spacing. The same total study time spread across days beats one block. The catch: it needs planning ahead, so it loses to whatever is urgent.
• Interleaving. Mixing problem types teaches you to recognise which method applies. The catch: performance during practice drops, which reads as “this is not working”.
• Dual coding. Words plus a relevant visual beats either alone — for everyone, not for visual learners. Decorative images do nothing.

Notice what all four have in common: they feel worse while you do them and work better afterwards. Matching material to a preferred style does the opposite, which is a large part of why it persists. A method that feels easy is one people keep, whatever the outcome — the same trap behind timer-based study rituals and behind the 21-day rule.

The boring bottom line

Learning styles are a real observation with a fictional instruction attached. People do have preferences; those preferences do not describe a processing system; and teaching to them has never reliably outperformed simply teaching well. The strongest evidence in favour comes from small studies on filtered samples, and its own authors say it is too weak to act on.

If you took a style test years ago and have been organising your studying around the result, you have not been sabotaged. You have been spending effort on the wrong variable. Move it to how you test yourself and how you space the work, and the label stops mattering — which is what most durable behaviour change looks like once the branding is stripped off. And if the real problem is that you are not opening the material at all, that is a different mechanism with a different fix.

Sources

• Fleming, N. D., & Mills, C. (1992). Not Another Inventory, Rather a Catalyst for Reflection. To Improve the Academy, 11(1), 137-155. doi:10.1002/j.2334-4822.1992.tb00213.x
• Pashler, H., McDaniel, M., Rohrer, D., & Bjork, R. (2008). Learning Styles: Concepts and Evidence. Psychological Science in the Public Interest, 9(3), 105-119. doi:10.1111/j.1539-6053.2009.01038.x PMID 26162104
• Newton, P. M., & Salvi, A. (2020). How Common Is Belief in the Learning Styles Neuromyth, and Does It Matter? A Pragmatic Systematic Review. Frontiers in Education, 5, 602451. doi:10.3389/feduc.2020.602451
• Clinton-Lisell, V., & Litzinger, C. (2024). Is it really a neuromyth? A meta-analysis of the learning styles matching hypothesis. Frontiers in Psychology, 15, 1428732. doi:10.3389/fpsyg.2024.1428732 PMID 39055994
• Hattie, J., & O’Leary, T. (2025). Learning Styles, Preferences, or Strategies? An Explanation for the Resurgence of Styles Across Many Meta-analyses. Educational Psychology Review, 37, 31. doi:10.1007/s10648-025-10002-w
• Cuevas, J., & Dawson, B. L. (2018). A test of two alternative cognitive processing models: Learning styles and dual coding. Theory and Research in Education, 16(1), 40-64. doi:10.1177/1477878517731450
• Husmann, P. R., & O’Loughlin, V. D. (2019). Another Nail in the Coffin for Learning Styles? Anatomical Sciences Education, 12(1), 6-19. doi:10.1002/ase.1777 PMID 29533532
• Kratzig, G. P., & Arbuthnott, K. D. (2006). Perceptual learning style and learning proficiency: A test of the hypothesis. Journal of Educational Psychology, 98(1), 238-246. doi:10.1037/0022-0663.98.1.238

Our search measurement was taken on a single day from one location, without control over personalization, and it describes what a search engine surfaced rather than a fixed ranking. Anyone repeating it will get a slightly different set of pages and should expect slightly different counts.

One email. One tested idea. No questionnaire required.

The rest of the citation trails we have followed: the Myths library.