The learning styles evidence in language teaching

A language teacher hears "learning styles are a myth" roughly once a year now, usually from someone waving the same 2008 review. She nods, then goes back to planning a lesson with a visual warmer, an audio gap-fill and a running dictation. That's what her training told her to do.

She isn't being stubborn. The debunking she keeps hearing was never about her subject.

So I checked what the evidence says about language teaching specifically. There's more of it than the myth-busting posts suggest, and it mostly comes from EFL (English as a foreign language) classrooms.

It's also carrying more weight than anyone has noticed.

Meshing needs a crossover, and the 2008 review found virtually none

Most popular debunkings garble this, so it's worth getting right. Pashler, McDaniel, Rohrer and Bjork named the specific claim in their 2008 review for Psychological Science in the Public Interest:

The most common…hypothesis about the instructional relevance of learning styles is the meshing hypothesis, according to which instruction is best provided in a format that matches the preferences of the learner (e.g., for a "visual learner," emphasizing visual presentation of information).

Pashler et al. 2008, p. 105

The test they specify is a crossover interaction, and the word crossover is doing all the work. Visual learners have to do better on visual instruction and auditory learners have to do better on auditory instruction.

Both arms have to bend, in opposite directions. If everyone does better with audio, that's a fact about audio, and it says nothing about learner types.

Plenty of studies claiming to support learning styles report exactly that, and the crossover requirement is what catches them.

Their verdict: "virtually no evidence for the interaction pattern", and of the studies using an appropriate method, "several found results that flatly contradict the popular meshing hypothesis."

Now the part I haven't seen written down. I went through every empirical study Pashler's team evaluated.

The domains: computer-based electronics lessons, internal medicine modules, free recall of words and line drawings, sixth-grade reading programmes, fourth-grade fractions, a letter-series task, and an introductory psychology course at Yale.

Not one is a foreign language study. The review that settled this question for education never touched the subject the teacher above actually teaches.

Five language studies give the biggest effects and almost no crossovers

That gap got filled by a strange route.

In July 2024, Virginia Clinton-Lisell and Christine Litzinger published a meta-analysis (a study that pools the results of other studies) in Frontiers in Psychology. It had an awkward title for the consensus: "Is it really a neuromyth?"

They pooled 21 studies, 101 effect sizes, 1,712 participants, and found a positive overall result. Hedges' g = 0.32, SE = 0.12, 95% CI [0.07, 0.57], p = 0.01.

That confidence interval (the range the true effect probably sits in) nearly touches zero, but it clears it. It's the strongest published result the matching hypothesis has.

So I read the study table. Five of the 21 are explicitly English as a foreign or second language: Rassaei 2018 and 2019, Kam 2020, Tadayonifar 2021, Mujtaba 2022. Two are in Language Teaching Research, a serious applied linguistics journal.

Language teaching is 23.8% of this literature and 16.1% of its participants, 276 people. Rassaei 2018 ran 62 learners. Tadayonifar ran 13.

Then I went through Table 2 by hand, applying the authors' own rule for a crossover: at least two style groups have to score higher under matched instruction. Here's what the language studies do.

Rassaei 2018, delayed production, Hedges' g by learner group

auditory group +2.58 Matched to audio glosses. An enormous positive effect.

visual group −1.47 Matched to textual glosses. Matching made them worse.

Both figures from Clinton-Lisell & Litzinger 2024, Table 2. Only one arm is positive, so this is not a crossover, and by Pashler's criterion it is not evidence for meshing. It is evidence that audio glosses beat text glosses for everyone.

That's the thing worth knowing, and I haven't seen anyone say it. The three largest language effects in the meta-analysis, Rassaei 2018, Rassaei 2019 and Mujtaba 2022, all have this shape.

One group posts an effect between g = 1.35 and 2.58. The other group goes negative, down to −1.48.

The authors mark crossover rows in bold, and none of those three is marked. Seven studies in Table 2 are: Chen & Sun 2012, Chui et al. 2021, Hazra et al. 2013, Kam et al. 2020, Kassaian 2007, Lehmann & Seufert 2020, and Tadayonifar et al. 2021. Two of those seven are labelled EFL in Table 1, and one of the two, Tadayonifar, ran 13 people.

So the headline is backwards from how it reads. Language teaching contributes the biggest numbers in the supportive meta-analysis and almost none of its actual support.

A learner sorted into the "wrong" group landed well below the alternative. That's the pattern you get if audio glosses (notes explaining a word, given as sound) simply beat text glosses and the labels are noise.

One detail makes that reading hard to avoid. Rassaei tested vocabulary in writing, and the audio-gloss learners still won by more than 2 standard deviations. A hearing-word advantage that survives onto a written test is a fact about hearing the word.

The meta-analysts are blunt about quality: "five studies were determined to WWC standards and the remainder did not meet WWC standards." Sixteen of 21 failed the What Works Clearinghouse bar.

Counting crossovers by outcome rather than by study, 11 of 42 measures showed the pattern, so 73.8% didn't. Their conclusion, which the headline effect size hides: the benefits are "too small and too infrequent to warrant widespread adoption."

Teacher-training pages still sell learning styles

Meanwhile, here's what a new teacher reads.

The TEFL Institute of Ireland publishes a page on teaching to learning styles, first published 19 January 2024 and updated 6 March 2026. It opens by telling trainees this is already part of their qualification:

If you have undertaken or are currently studying a TEFL course, you will have learned all about the different TEFL Strategies and learning styles of students in the classroom.

TEFL Institute of Ireland, retrieved 3 September 2026

TEFL UK, in a post dated 18 November 2024, runs the four VARK categories: visual, aural, read/write and kinesthetic. It says adapting to them "is key to unlocking each student's potential." Visual learners get flashcards and pictures for new vocabulary; auditory learners get songs and repetition aloud.

The British Council is the interesting one, because it publishes both sides and doesn't appear to know it. Its TeachingEnglish reference entry on VAK tells teachers:

Learners can be encouraged to think about their learning style and how it affects their work by trying a quiz to identify their preferences.

British Council TeachingEnglish, VAK entry, retrieved 3 September 2026

Elsewhere on the same site, Sangeeta Sathe writes that "there is little to no improvement in learning outcomes when educators adapt their teaching to cater to learning styles".

Sathe warns about a learner who decides "they don't have to pay attention in reading and writing activities" because a quiz called them auditory.

One organisation, two pages, opposite advice, no cross-reference between them.

Preply, which sells language tutoring, published a learning styles study in April 2025 and updated it in April 2026. It tells readers "identifying your learning style could be the solution", maps US states to learner types using Google Trends search volume, then assigns study techniques by type. Spaced repetition is filed under "Best for: Kinesthetic learners, intrapersonal learners."

That one is the real cost. Spaced repetition (reviewing at gradually longer gaps) is among the best-evidenced findings in the psychology of memory, and it works for everybody. Filing it under two personality types tells everyone else to skip it.

Credit where it belongs: Berlitz, a language school with something to lose, published a piece in September 2022 titled "Does the learning styles theory apply to language learning?"

It answered honestly, concluding that VARK "is best applied as a multi-modal style rather than an individual style."

The belief numbers explain why the rest keep publishing. Newton and Salvi pooled 37 samples covering 15,405 educators across 18 countries: 89.1% believed matching works, 79.7% said they use it.

Trainee teachers believed at 95.4%, higher than the 87.8% of qualified ones. The belief arrives during training and reaches you as advice about how to take notes in a lesson long before anyone mentions evidence.

Sound and image help everyone, and matching by type doesn’t

Here's where language teaching earns a real exception, and where a lot of debunking overshoots.

You can't learn Spanish pronunciation from a page. In this subject, audio is where half the content lives.

A written flashcard for ciudad without sound is missing the thing you need to say it. That's a fact about the material, whoever you are.

Manuela Macedonia made this argument in 2015 in Frontiers in Psychology. It's the paper the general debunkings never cite, because it's about vocabulary rather than schools.

She says there's "no empirical evidence for the existence of learning styles", then explains what the brain does instead:

The natural way for the brain to learn words is by collecting multiple sensory and sensorimotor experiences…Impoverished linguistic input by allowing only one modality, for example only acoustic or visual input, the so called learning style of the student, reduces the chances of acquiring words.

Macedonia 2015, Frontiers in Psychology 6:1800

Read that carefully, because it inverts the practical advice. Teaching a learner in one modality because a quiz labelled them strips a channel out of the input.

Macedonia goes further: "it is possible that learning according to one's presumed learning style hinders learning."

She also names the measurement problem. Learning styles are almost always established by self-report (a questionnaire the learner fills in), where "subjects might lack introspection" and answers reflect what feels familiar rather than what works.

Rogowsky, Calhoun and Tallal ran the cleanest test of the modality version in 2015, following Pashler's design exactly. They sorted 121 adults, then compared audiobook against e-text, testing comprehension immediately and after 2 weeks.

No interaction anywhere: F(1, 37) = 0.54 on the delayed test.

The detail worth keeping is the shape of the graph. Both groups scored higher on listening than reading. And the people classified as visual learners beat the auditory group at listening comprehension, F(1, 66) = 5.48, p < .05.

The labels ran backwards.

So two claims that sound alike come apart.

Presenting a word as sound and image for everyone is dual coding, and it holds up: two traces, two retrieval routes. Presenting a word as sound only, for the learner a quiz called auditory, is meshing, and it doesn't.

One adds a channel, the other removes one. That's also why hearing a word before you read it is worth doing whichever type you think you are.

On Monday, keep the variety and drop the sorting

Stop taking the quiz. It measures what you're used to.

Keep every modality you were going to use anyway, and apply them to everything instead of sorting them by person. Hear the word, see the thing, say it aloud.

The production effect (saying a word aloud helps you remember it) is real for everyone, and so is the image.

If you're a teacher, the British Council's own line is the one to keep: vary the input, but not because it caters to every learning style. Vary it because that variety is what makes it stick.

And if you've been skipping listening practice because you decided years ago that you're a visual learner, that came from a questionnaire never shown to predict anything.

Rogowsky's data suggests the label pointed the wrong way anyway.

This is roughly the bet Verbamor makes. Every card carries the word, an image, and native audio of the example sentence, for every learner. The evidence supports stacking modalities rather than choosing between them.

No style quiz, and there won't be one. It's $19.99/month, iOS only, and it holds vocabulary from your own lessons rather than teaching you anything new.

If you'd rather build that yourself, Anki does images and audio on a card. The computer version is free, and AnkiMobile on iOS is a one-time US$24.99. A deck you maintain beats a subscription you abandon.

And if you've been differentiating by learning style for a decade, the useful half of what you built is the variety. Keep the activities. Drop the sorting.

Sources

The primary literature

Pashler, H., McDaniel, M., Rohrer, D., & Bjork, R. (2008). Learning styles: Concepts and evidence. Psychological Science in the Public Interest, 9(3), 105–119. Full text read. Meshing definition p. 105; "virtually no evidence for the interaction pattern" and "flatly contradict" in the summary; the crossover criterion in "Interactions as the Key Test." Studies evaluated: Sternberg et al. 1999 (psychology course), Massa & Mayer 2006 (electronics lesson), Cook et al. 2009 (internal medicine), Constantinidou & Baker 2002 (free recall), plus the ATI literature (Freebody & Tirre 1985 reading, Cramer et al. 1989 fractions, Greene 1980 letter-series). No second language study appears in the review.
Macedonia, M. (2015). Learning styles and vocabulary acquisition in second language: How the brain learns. Frontiers in Psychology, 6:1800. doi:10.3389/fpsyg.2015.01800. Quotes verbatim: "no empirical evidence for the existence of learning styles"; the multiple-modalities passage; "it is possible that learning according to one's presumed learning style hinders learning"; the self-report critique.
Rogowsky, B. A., Calhoun, B. M., & Tallal, P. (2015). Matching learning style to instructional method: Effects on comprehension. Journal of Educational Psychology, 107(1). Full PDF read. N = 121 classified, 61 in the instructional experiment; audiobook vs e-text; no interaction immediately or at 2 weeks, delayed F(1, 37) = 0.54, p > .05. Visual-word group outscored the auditory group on listening comprehension, F(1, 66) = 5.48, p < .05.

The meta-analysis and the arithmetic on it

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. Read as the published PDF, Tables 1 and 2 in full. 21 studies, 101 effect sizes, 1,712 participants; g = 0.32, SE = 0.12, 95% CI [0.07, 0.57], p = 0.01. Verbatim: "five studies were determined to WWC standards and the remainder did not meet WWC standards"; 11 of 42 outcome measures showed crossover, "26.19%"; conclusion "too small and too infrequent to warrant widespread adoption." Their crossover rule, verbatim: "Learning outcomes indicating a crossover interaction as articulated in Pashler et al. (2008) in which at least two styles had higher learning outcomes with matched instruction are bolded in Table 2."
The second language studies inside it, from Table 1: Rassaei 2018 (N = 62, EFL, textual vs audio glosses, Language Teaching Research 22(6) 657–675, doi:10.1177/1362168817690183); Rassaei 2019 (N = 61, EFL, corrective feedback on English articles); Kam et al. 2020 (N = 60, EFL, video lecture with and without captions); Tadayonifar et al. 2021 (N = 13, EFL, text vs audio glosses); Mujtaba et al. 2022 (N = 80, ESL, audio vs text grammar instruction). Table 1 records Rassaei 2018's outcome as "vocabulary production and recognition (written)", which is why the audio advantage there cannot be a test-format artefact. Table 2 values: Rassaei 2018 auditory +1.88 to +2.58, visual −1.23 to −1.48; Rassaei 2019 auditory +1.59 to +1.77, read/write −0.04 to −0.34; Mujtaba 2022 auditory +1.35 to +1.73, visual −0.53 to −0.63.
My arithmetic on those tables. 5 of 21 studies = 23.8% second language; 276 of 1,712 participants = 16.1%. The authors name the seven crossover studies outright in the discussion: "approximately one-fourth of the learning measures indicated a crossover interaction in which there were positive effect sizes for matched instruction for two different learning styles (Kassaian, 2007; Chen and Sun, 2012; Hazra et al., 2013; Kam et al., 2020; Lehmann and Seufert, 2020; Chui et al., 2021; Tadayonifar et al., 2021)." That list also matches the bolded rows in Table 2, so the two agree. Two of those seven are explicitly labelled EFL in Table 1. Note that Moser & Zumbach 2018 and Riding & Douglas 1993 each have two style groups above zero on some measure but are not bolded, so a naive g > 0 reading of the table overcounts; the bolding is the authors' judgement and I have deferred to it. Rassaei 2018, Rassaei 2019 and Mujtaba 2022, the three largest language effects in the pool, each have one group positive and the other negative, so none is a crossover. 42 − 11 = 31 of 42 outcome measures, 73.8%, did not show the pattern; 21 − 5 = 16 studies, 76.2%, failed the WWC screen. The authors print none of these breakdowns. The seven-study crossover list is theirs, quoted above from their discussion.
Kassaian 2007 is described in Table 1 as 66 university students learning "new vocabulary words," which is very likely an EFL sample, but the meta-analysis does not label it EFL and I could not reach the original paper. I have not counted it as a language study, which makes my language share of the crossover set a conservative floor rather than a ceiling.
Newton, P. M., & Salvi, A. (2020). How common is belief in the learning styles neuromyth, and does it matter? Frontiers in Education, 5:602451. 37 samples, 15,405 educators, 18 countries. Verbatim: belief "with a weighted percentage of 89.1%, ranging from 58 to 97.6%"; use or planned use 79.7%; trainee teachers 95.4% vs qualified 87.8%.

What language teaching publishes, retrieved 3 September 2026

TEFL Institute of Ireland. TEFL strategies for teaching different learning styles. Published 19 January 2024, last updated 6 March 2026, by Rachel Hammond. Quote verbatim as printed. Gives per-style activity lists for visual, auditory, kinesthetic, social and solitary learners.
TEFL UK. A guide to different learning styles in TEFL. Dated 18 November 2024. Runs the four VARK categories; "Understanding these different styles and adapting your teaching methodology accordingly is key to unlocking each student's potential."
British Council TeachingEnglish. Visual, auditory and kinaesthetic (VAK). No visible publication date. Quote verbatim; the entry carries no caveat about the evidence and does not link to the article below.
Sathe, S. British Council TeachingEnglish. The truth behind learning styles. Quotes verbatim, including the recommendation to vary tasks "NOT because it caters to every learning style but because this variety in learning is what makes it stick."
Preply. What are the most common learning styles and which one is preferred? By Anna Pyshna. datePublished 2025-04-03, dateModified 2026-04-22 in the page's own JSON-LD. Assigns spaced repetition to "Best for: Kinesthetic learners, intrapersonal learners." Methodology section states the state-by-state learner types were derived from Google Trends search volume.
Briggs, S. Berlitz. Does the learning styles theory apply to language learning? Dated 8 September 2022. Concludes VARK "is best applied as a multi-modal style rather than an individual style", and quotes Macedonia and Newton & Salvi accurately.

What I could not verify

The five second language studies were read through the meta-analysis's own tables, not as full texts. Rassaei 2018's publisher page returns HTTP 403 to automated fetching and the other four sit behind paywalls. The effect sizes and sample sizes above are Clinton-Lisell and Litzinger's coding of those papers. If their coding is wrong, everything here moves with it. The crossover classification is mine, computed from their published numbers by their stated rule.
Countries of origin for the five studies are not given in the meta-analysis, so I have not named any, though the author names and EFL context suggest Iran for the Rassaei and Tadayonifar samples. I have not asserted that in the post.
The meta-analysis reports 11 bolded outcome measures; the font extraction gives 7 bolded studies. Those count different things, so both hold, but I have not reconstructed their per-measure tally row by row.
I found no published critique of learning styles written specifically for the language teaching profession other than Macedonia 2015. If one exists in ELT Journal or Language Teaching, I did not find it.