Are Learning Styles Debunked? What the Research Actually Shows

Most people who build courses have met learning styles somewhere: a questionnaire in teacher training, a “visual, auditory or kinaesthetic?” slide in an onboarding module, a client asking for “something for the hands-on learners”. The idea is simple and appealing. Each of us has a style, and we learn best when teaching matches it.

The short answer to whether learning styles are debunked is: the matching claim is, as far as the evidence can show. People do have preferences, and some people are better with pictures or words than others. What decades of research haven’t found is the thing that would matter to a course designer: that teaching each learner in their preferred style makes them learn more.

This entry covers what the research tested, what the newest meta-analysis adds, why the idea survives, and what to design around instead.

What the research says

Many models, one claim

There isn’t one learning styles theory. A review for the UK’s Learning and Skills Research Centre identified 71 models of learning styles and examined 13 of the most influential in detail (Coffield et al., 2004). The best known in online learning are the sensory ones: visual, auditory and kinaesthetic (VAK), and VARK, which adds read/write.

What they share is a practical promise. Pashler, McDaniel, Rohrer and Bjork called it the meshing hypothesis: instruction works best when its format matches the learner’s preferred style (Pashler et al., 2008). That’s the claim worth testing, because it’s the one that would change how you build a course.

The test the theory has to pass

Pashler and colleagues’ main contribution was to spell out what evidence for matching would look like. You classify learners by style, randomly assign learners of each style to different teaching methods, then give everyone the same test. Matching is supported only if the results cross over: visual learners do best with the visual method, and auditory learners do best with the auditory method.

That design rules out the two results most often offered as evidence. “Learners liked the lesson in their style” is about preference, not learning. “A lesson with pictures and narration beat a lesson with narration alone” is a method that helps everyone, which says nothing about matching.

When the authors searched the literature, they found plenty of evidence that people state preferences, but very few studies with a design that could test the meshing hypothesis, and several well-designed ones whose results contradicted it. They also noted that many learning styles models have never been tested this way at all, so “unsupported” is more accurate than “disproved” for those. Their conclusion was blunt.

Experiments since then

The clearest follow-ups come from Beth Rogowsky and colleagues, who ran the experiment Pashler described. Their 2015 study with adults, the first to follow Pashler’s design, tested whether a visual or auditory preference predicts comprehension depending on how material is taught. Their 2020 paper repeated the design with fifth graders at a school in Pennsylvania, using a learning styles inventory to classify students as visual or auditory, and having each student experience both listening and reading on different days (Rogowsky et al., 2020).

In both, the results failed to show the crossover. In the 2020 study, there was no significant relationship between auditory or visual preference and comprehension. Students classified as visual scored higher than those classified as auditory on both the listening and the reading measures, which is a difference between groups of learners, not a benefit of matching. The authors describe their earlier adult experiments in the same terms: “Taken together, these experiments found no evidence to support learning styles-based instruction.”

What learners feel versus what they learn

One reason the idea feels right is that preferences do shape experience. Knoll and colleagues gave 52 undergraduates a questionnaire on whether they think in words or pictures, then had them study pairs of pictures and pairs of words and predict how well they would remember them (Knoll et al., 2017). Learning style didn’t predict recall of either kind of material. It did predict learners’ immediate judgements: self-described verbalisers expected to remember the words better, and self-described visualisers the pictures.

It was a small study with one group of students, but it fits the pattern. Matching can make a lesson feel easier and more comfortable. That isn’t the same as learning more from it, and course evaluations that ask “did this suit your learning style?” are measuring the first.

The 2024 meta-analysis

The most recent summary is a meta-analysis by Virginia Clinton-Lisell and Christine Litzinger, titled “Is it really a neuromyth?” They searched for studies that matched instruction to modality learning styles (visual, auditory, kinaesthetic and similar, often measured with VARK) and found 21 eligible studies with 101 effect sizes and 1,712 participants (Clinton-Lisell & Litzinger, 2024).

Matched instruction did come out ahead on average, with an effect size of g = 0.31 (95% confidence interval 0.05 to 0.57). That’s the number some summaries quote as a revival of learning styles. Read on, though:

• The crossover was rare. Only 26% of learning outcome measures showed matched instruction helping at least two styles, the pattern the theory actually predicts. In 12 further studies that lacked the statistics for the main analysis, 25% showed it.
• The studies were weak. Five of the 21 met the What Works Clearinghouse’s standards for research design; the rest did not. The authors also included quasi-experiments, which don’t randomly assign learners to conditions.
• The interval is wide. A true effect anywhere from close to zero to moderate fits the data.

The authors’ own reading is cautious.

So the newest evidence doesn’t rescue learning styles. It suggests that, under some conditions, some matching may do a little good, and that finding those conditions would take better studies than we have.

Why the idea survives

Belief in matching remains common among educators. A survey of 242 teachers in the UK and the Netherlands who were interested in the neuroscience of learning found 93% of the UK teachers and 96% of the Dutch teachers agreed that people learn better in their preferred learning style (Dekker et al., 2012). Teachers with more general knowledge about the brain were more likely, not less, to believe common neuromyths.

A later systematic review pooled 37 studies of 15,405 educators in 18 countries. On average, 89.1% endorsed matching instruction to learning styles, ranging from 58% to 97.6%, with no evidence that belief had declined in recent years (Newton & Salvi, 2020). The authors also note the surveys’ own limits, such as convenience samples and loosely worded questions.

Belief and use aren’t the same. In a survey of 114 UK university academics, 58% believed in matching, but 33% reported actually using learning styles in the previous 12 months (Newton & Miah, 2017). Before reading a summary of the evidence, 64% said they try to organise their teaching to accommodate learning styles; afterwards, about 32% said they planned to.

How strong is the evidence?

Strong against matching as a design rule. The best-designed experiments don’t find the crossover, the most recent meta-analysis finds only a small and infrequent benefit from low-quality studies, and the reviews agree. Two limits are worth keeping in view. Many of the 71 models have never been tested properly, and most experiments test short tasks such as reading comprehension or recall rather than whole courses. In practice, we’d treat a learner’s stated style as useful information about what they enjoy, and nothing more.

How to apply it in a course

The useful question isn’t “what style is each learner?” but “what format does this content need, and what does this learner already know?”

We’ll use one invented example: a self-paced course teaching new hospitality staff to make espresso drinks, with a client who has asked for “visual, audio and hands-on versions” of each lesson.

1. Don’t sort learners, and don’t build parallel versions

Drop the learning styles questionnaire from onboarding, and don’t build three versions of each lesson. On the evidence, the extra versions cost a great deal of production time for a benefit that may not exist. In the espresso course, that’s a single well-made lesson per skill instead of three weaker ones.

2. Choose the format the content needs

Format should follow what’s being taught, for every learner. Grinding and tamping is a physical skill, so it needs a close-up demonstration and then practice on the machine. Milk texture is partly about sound, so the lesson needs audio of the steam wand at each stage. The menu and its ratios are a structure, so a diagram does the work. Where words and pictures can work together, use both: the evidence on combining narration with relevant visuals is far stronger than anything for learning styles, and we cover it in our entry on dual coding.

3. Adjust for prior knowledge, not style

The individual difference that does change what works is how much learners already know. A barista with two years’ experience needs less step-by-step guidance than a first-week starter, and detailed guidance that helps a novice can slow an expert down. Our entry on cognitive load theory explains why. Offer a short diagnostic quiz and let experienced learners skip ahead, rather than a styles questionnaire.

4. Offer access options to everyone

Captions, transcripts and downloadable recipe cards are worth providing, but for access, not styles: for learners who are hard of hearing, working in a noisy café, on a poor connection, or reading in a second language. They’re cheap to add and help whoever uses them.

5. Spend the saved effort on practice

The time you don’t spend on parallel versions goes further on retrieval and feedback. End each lesson with questions that make learners recall the ratios and steps, write feedback that tells them what to change (our entry on effective feedback covers how), and bring the key points back a week later, as described in our entry on microlearning.

6. Answer the client with the evidence

When a client or manager asks for learning styles, explain what the research found and offer what it does support: formats chosen for the content, access options for everyone, and more practice. Most people asking for “something for the hands-on learners” want the course to involve doing, which is a good instinct for a practical skill.

This fits the design phase of the ADDIE model, when you choose formats for each module. Our online course design checklist includes the check that every format choice is made for the content, not for a style.

How the platforms support it

Moving away from learning styles asks less of a platform, not more: no learner profiling, no branching by style. What matters is whether you can put the right format in each lesson and offer access options. We haven’t yet published our reviews of Thinkific, LearnWorlds, Teachable and Kajabi, and features change often, so check these in your platform’s current documentation, on the plan you’re paying for:

• Mixed media in one lesson. Can a single lesson hold video, text, images and audio together, or does each format need its own lesson?
• Captions and transcripts. Can you upload caption files or a transcript for each video, and can learners turn captions on in the player?
• Downloads. Can you attach a PDF or image to a lesson for learners to keep?
• Skipping ahead. Can learners who pass a diagnostic quiz skip lessons, or is every course locked to a fixed order?
• Quizzes in every module. Can you add a short quiz to any lesson, not only at the end of the course?

If your platform locks the order, a clear “If you already know this, go to lesson 4” note in the lesson text does much the same job.

Further reading

• Pashler, McDaniel, Rohrer & Bjork (2008), “Learning styles: Concepts and evidence”. The review that defined the test learning styles have to pass. Its explanation of the crossover design is the most useful part for anyone judging a claim.
• Clinton-Lisell & Litzinger (2024), “Is it really a neuromyth?”. The newest meta-analysis, open access. Worth reading in full before quoting its headline effect size.
• Rogowsky, Calhoun & Tallal (2020), “Providing instruction based on students’ learning style preferences does not improve learning”. A clear example of the right experiment, run with school students. Open access.
• Newton & Salvi (2020), “How common is belief in the learning styles neuromyth, and does it matter?”. A systematic review of the surveys, with a thoughtful discussion of whether belief alone does harm. Open access.
• Coffield et al. (2004), “Learning styles and pedagogy in post-16 learning”. Long, but the best map of the 71 models and where each one came from.

For the rest of the principles behind good course design, browse our research library.