Do Mind Maps Work? What the Research Says

Do Mind Maps Work? The Short Answer

Yes, with two caveats that change the answer a lot. Mapping produces a moderate benefit in the research literature, around g = 0.58 in the largest meta-analysis, but nearly all of that advantage comes from comparing mapping against passive study. When the control group writes its own outline instead, the advantage drops to roughly a tenth of that.

The second caveat is bigger. Almost all of this evidence is about concept maps, not the radial mind maps most people picture. Those are different formats with different rules, and the studies that gave us the famous numbers tested the other one.

So the useful question isn't "do mind maps work." It's "what am I replacing, and with which format?" Swap a lecture for a map and you'll learn more. Swap your notes for a map and the research says you'll land in about the same place, more slowly.

---

Mind Maps and Concept Maps Are Not the Same Thing

People use the terms interchangeably. The research doesn't, and the difference explains a lot of the conflicting results.

Joseph Novak developed the concept map at Cornell in 1972, while running a longitudinal study of how children's understanding of science changed over time. He built it on David Ausubel's assimilation theory, which holds that you learn by hooking new ideas onto ideas you already have. The defining feature is the labeled linking phrase: words written on the line between two concepts. A concept map doesn't say that "funding" and "documentary" are related. It says "funding constrains documentary," and that's a claim you can be wrong about.

Tony Buzan popularized the mind map in 1974 through the BBC series and book Use Your Head. It's radial. One topic sits in the center, branches grow outward, each branch carries a single keyword, and color and imagery do a lot of work. Buzan's rules do allow arrows between branches, so cross-connection isn't forbidden, but the lines carry keywords rather than relationship labels and the underlying shape is a tree.

That "can it be wrong" row is the one that matters. A structure that can be wrong is a structure you can check, and checking is where learning happens.

One more thing worth flagging: Buzan's original justification leaned on claims about hemispheric specialization and "whole-brain" processing. Those haven't held up. A 2013 imaging study of 1,011 people by Nielsen and colleagues found no evidence for the left-brained or right-brained network phenotype the popular version assumes. That's the same family of reasoning behind the learning styles myth, which is still taught despite failing every controlled test. A weak explanation doesn't make a technique useless. It does mean you should look at outcomes rather than at the story.

---

What the Mind Mapping Research Actually Measured

Two large syntheses anchor this literature, and they were written by overlapping teams.

Nesbit and Adesope (2006), published in Review of Educational Research, pulled 67 effect sizes from 55 studies covering 5,818 participants, from Grade 4 through university, across science, psychology, statistics and nursing.

Schroeder, Nesbit, Anguiano and Adesope (2018), in Educational Psychology Review, extended that work to 142 independent effect sizes and 11,814 learners. Their random-effects model returned g = 0.58 overall. Split by activity, creating a map returned g = 0.72 and studying a pre-made map returned g = 0.43. Because the newer paper extends the older one, you can't add the two samples together, and people quoting a combined figure are double-counting the same students.

Now the part that almost never survives the trip into a blog post. Search the 2006 meta-analysis for the word "mind" and you get nothing. Not once. Neither "Buzan" nor "radiant" appears either. Both papers define their subject as concept and knowledge maps, meaning node-link diagrams whose links identify the relationship between concepts. That's precisely the feature a Buzan mind map doesn't have.

So when you see "studies show mind maps improve learning by 58%," two things have gone wrong. The number is a standardized mean difference, not a percentage. And it came from a different diagram. The evidence base for radial mind maps specifically is much thinner, and we'll get to the two trials that make it up.

---

Do Mind Maps Beat Note-Taking? The Comparison Problem

An effect size is a comparison, and a comparison has two sides. Nesbit and Adesope broke their data out by what the control group was actually doing. Here are the main rows of their Table 6, dropping two categories that rested on a single study each.

Read the first two rows together. The same activity, measured in the same meta-analysis, produces an effect nearly four times larger when the thing it replaces is passive. Building a map outperforms a lecture comfortably. Building a map instead of writing an outline yourself isn't much of anything.

The authors don't hedge about this. In their words, concept mapping "appears to compare very favorably with teaching methods in which learners have diffuse responsibility for task completion (e.g., whole-class discussion), but it shows only a small advantage over other constructive tasks, such as individual note-taking or summarizing. As an effect size drops below .2 standard deviations, one may be justified in questioning its pedagogical significance and whether it might be attributed solely to experimenter bias."

They go further. Of the 10 studies using mixed group and individual mapping, 8 used lecture or discussion as the comparison, and the authors conjecture that many of those elevated effect sizes are "due to lower effectiveness of the comparison treatment rather than to any particular benefit" of mixed group and individual concept mapping.

This is the single most important fact about mapping research, and it's almost never in the blog posts. The technique isn't magic. Doing something is better than doing nothing, and mapping is a something. So is writing.

Two other moderators are worth knowing, with their limits attached:

• Study length points in opposite directions across the two papers. The 2006 analysis of map construction found larger effects in studies running five weeks or less (g = 0.701) than in longer ones (g = 0.363). The 2018 analysis found the reverse, with the weakest effects in the shortest studies. When two overlapping teams get opposite signs, nobody should be quoting either as settled.
• The instruction may matter more than the tool. In one study the authors reviewed, by Patterson, Dansereau and Newbern, pairs of students given an explicit procedure for using their maps gained g = 0.65, while pairs handed the same maps with no strategy gained g = 0.29. That's a single experiment quoted as an illustration, not a moderator analysis, so treat it as a hypothesis worth holding.

Age does something strange too. In the construction studies, Grades 4 to 8 returned g = 0.905 and university students g = 0.773, but Grades 9 to 12 returned just g = 0.165. Nobody has a clean explanation for the dip.

---

Who Concept Maps Help: The Finding Almost Nobody Quotes

Nesbit and Adesope also split the "studying a ready-made map" studies by learner ability. These are small sets, the ability measures were relative median splits rather than standardized tests, and only two rows reach statistical significance. Read them as a direction, not a result.

The pattern: learners with less to work with gain from being handed a structure, and learners with more don't show a measurable gain. The negative point estimates come from the verbal ability split, on two studies and 86 people. The rows that measure prior knowledge of the subject sit at roughly zero and slightly positive.

The authors block the obvious over-reading themselves: "Because the statistical power of these comparisons is low and the confidence intervals are wide, it cannot be concluded that maps do not benefit higher-ability students." That sentence belongs next to the table every time the table gets quoted, so here it is.

What you can take from it is modest and still useful. A finished map is scaffolding. If you don't yet have a structure for the material, someone else's structure helps a lot. If you already have one, the evidence that it helps you further is absent rather than negative. One plausible mechanism is the expertise reversal effect, where support that assists a novice starts interfering once the learner has their own schema, described by Kalyuga and colleagues in 2003. That's an interpretation we're proposing, not something either meta-analysis tested, and we covered the underlying idea in cognitive load theory for readers.

The practical version: if you're new to a topic, a good map someone else made is genuinely worth studying. If you're not, build your own instead of collecting theirs.

---

Two Medical School Trials, Two Different Answers

Meta-analyses smooth over individual studies. Two trials in medical education tested mind maps specifically, in the population that evangelizes them hardest, and they didn't agree.

Farrand, Hussain and Hennessy (2002), Medical Education 36:426-431, gave 50 students a 600-word text and had them either mind map it or use whatever study method they normally preferred. At one week, adjusted for baseline, the mind map group recalled 10.7% more than the controls. The confidence interval ran from -1.1% to 22.5%, p = 0.07, so on that comparison the result doesn't clear the usual bar.

There's a twist the summaries mangle. The mind map group's motivation was significantly lower than the controls' (2.8 versus 3.2, p = 0.02). The authors treated that as a confound suppressing the effect rather than as a strike against the technique, and when they adjusted for it the gap widened to 15.3%, CI 3.3% to 27.3%, p = 0.013. Their conclusion was that mind maps "provide an effective study technique when applied to written material." Note what that implies for you: the benefit showed up in students who didn't enjoy doing it.

D'Antoni and colleagues (2010), BMC Medical Education 10:61, randomly assigned 131 first-year medical students to standard note-taking (65 students) or mind mapping (66 students).

Nothing. The authors concluded that "mind mapping was not found to increase short-term recall of domain-based information or critical thinking compared to SNT," while noting that novices who'd just learned the technique weren't put at a disadvantage either. Their own main caveat is that the mind map group's entire training was a single 30-minute presentation, which is not much preparation for a technique that takes practice.

So: one small trial where mind maps probably helped, one larger trial where they clearly didn't, and a large concept-map literature saying the benefit over real note-taking is small. That's the honest state of the evidence, and it's a long way from the marketing.

---

What Probably Makes a Map Work

Here the confidence has to drop, because the meta-analyses measured whether maps help, not which parts of them do. Neither paper tested link labeling against no link labeling, and Schroeder's team reported no differences among the map types they could distinguish. What follows is the best reading of the theory, flagged as such.

The link is the plausible active ingredient. Novak's insight was never the diagram. It was that you have to name the relationship. "Inflation" next to "interest rates" is a doodle. "Inflation triggers interest rate rises" is a proposition you can check, defend, or discover is backwards. Writing a linking phrase makes you retrieve and self-explain at the same time, which are two of the better-evidenced things in learning science. That the concept-map literature is stronger than the mind-map literature is consistent with this, though consistency isn't proof.

Cross-links are where the work is. Connecting two branches that started in different parts of the map is the hardest move and the best sign you understand the material. A map with no cross-links is an outline drawn sideways.

Generating probably beats receiving, with an asterisk. Schroeder's split shows creating at g = 0.72 and studying at g = 0.43. But those are measured against their respective control conditions, and creation studies leaned more on lecture controls. Having just spent a section showing that the control drives the number, we shouldn't treat 0.72 versus 0.43 as a clean head-to-head. The broader case for effortful generation rests on desirable difficulties instead, which is better evidenced than anything in the mapping literature.

Dual coding is a bonus, not the engine. Paivio's theory says verbal and visuospatial knowledge sit in separate but linkable memory codes, so spatial layout can add a second retrieval route. Real, and we wrote about it in dual coding. It doesn't require a radial layout, and it isn't large enough to carry the results by itself.

Notice what's absent: color coding, branch thickness, central images, one word per line. Buzan's rules are mostly untested at the level of the individual rule. They aren't harmful. They just aren't the reason the technique works when it works.

A quick diagnostic. Look at a map you've made and count the lines that have a relationship written on them. If that's close to zero, you drew a table of contents in a circle, and a table of contents is not a model.

---

Build the Map From Highlights, Not From a Blank Page

In practice, the failure mode usually isn't a bad map. It's no map, because starting from a blank canvas after you've finished reading takes real activation energy, and by then you've forgotten which parts mattered.

Working from what you already marked fixes that. The passages you highlighted while reading are your candidate nodes, so mapping stops being "recall the whole chapter" and becomes "arrange these twenty things and explain how they touch."

A workflow that fits the evidence:

1. Highlight while reading, sparingly. With Glasp's web highlighter you mark passages in place without leaving the page. Fewer, better highlights beat a yellow page, for the reasons in the science of highlighting.
2. Wait. Don't map immediately. A day's delay turns the exercise into retrieval instead of copying.
3. Lay out only the nodes. Pull your highlights into a working space and arrange them spatially. No lines yet.
4. Draw lines and label every one. This is the step. If you can't name a relationship, you've found a gap, which is the point of the exercise rather than a failure of it.
5. Force two cross-links. Connect branches that don't obviously belong together.
6. Close everything and redraw from memory. The second map, drawn blind, is the one that tells you what you actually know.

Step 6 is non-negotiable. Skip it and you're admiring your own notes and calling the warm feeling comprehension, which is the illusion of competence doing exactly what it does.

The same loop runs on other sources. Video lectures collapse into mappable nodes through YouTube Summary, which pulls out timestamped key points you can treat as concepts. Book arguments come across through Kindle highlights, where passages you marked months ago are already isolated.

One caution follows from everything above: name your links yourself before you look at anyone else's. Once you've committed to a relationship, Glasp's AI chat is useful for the opposite job, arguing the reverse case so you have something to test your link against. Asking it to hand you the relationships in the first place removes the only step that reliably does the work.

---

When to Skip the Mind Map Entirely

Mapping is a relationship tool. When the thing you're learning doesn't have interesting relationships, it's overhead.

The prose row is the one people resist. If you know a topic well, writing a plain paragraph explaining it exposes gaps faster than a diagram will, because prose forces you to commit to an order and to transitions. That's the Feynman technique, and it costs nothing.

If you're choosing between mapping and other note formats generally, the comparison is closer than the internet suggests. Cornell notes and the other structured approaches in our note-taking methods guide land in the same broad performance band. Pick for the shape of the material, not the shape of the tool.

---

Frequently Asked Questions

Do mind maps actually improve memory?

The honest answer is that most of the evidence people cite is about concept maps, not mind maps. Schroeder and colleagues found an overall effect of g = 0.58 across 142 effect sizes for concept and knowledge maps. Nesbit and Adesope showed that when the comparison group wrote its own text or outline, the advantage fell to g = 0.194. For radial Buzan-style mind maps specifically, the direct evidence is two small medical school trials that disagree.

What is the difference between a mind map and a concept map?

A mind map is radial, with one central topic and branches carrying single keywords, popularized by Tony Buzan in 1974. A concept map is a network with labeled linking phrases stating the relationship between concepts, developed by Joseph Novak at Cornell in 1972, with cross-links between branches as a core feature. Nearly all the published effect sizes come from concept maps.

Are mind maps better than taking normal notes?

The evidence says not meaningfully. Nesbit and Adesope measured g = 0.194 for maps versus self-written text or outlines, and noted that below 0.2 standard deviations you can reasonably question the pedagogical significance. D'Antoni's 2010 trial of 131 medical students found no significant difference in recall between mind mapping and standard note-taking.

What are the disadvantages of mind mapping?

Three show up in the research. It takes longer than writing for a benefit that's small against active note-taking. Students in Farrand's trial were significantly less motivated using mind maps than their own preferred methods. And the format suits relationships, so it wastes effort on vocabulary, paired facts, or any linear procedure, where a list or spaced retrieval does more in less time.

Do mind maps work for everyone?

Probably not equally. When students studied a ready-made map, low verbal ability learners gained g = 0.436 while high verbal ability learners showed no detectable benefit. Those samples are small, and the authors caution that you cannot conclude maps fail to help stronger students. Building your own map is a different activity, and the same meta-analysis was unable to measure ability effects for it at all.

Are handwritten mind maps better than digital ones?

Neither meta-analysis reported the medium as a decisive moderator. Paper is faster for a first pass. Digital is better if you want to rearrange nodes, which matters because rearranging is how you find cross-links. Start from highlights you've already collected either way, so you aren't facing an empty canvas.

---

Do Mind Maps Work? The Honest Version

Mapping isn't a myth and it isn't a miracle. It's a decent tool for a specific job: making the relationships between ideas explicit enough that you can tell whether you actually understand them. Measured against passive study, it wins clearly. Measured against a student who's already writing thoughtful notes, it mostly ties.

That's a perfectly good result. It just doesn't support the claims on the cover of the book, and it mostly isn't even about the diagram on the cover of the book.

What survives every cut of the data is narrower and more useful than "draw a map." Generate the structure yourself rather than adopting one. Name every relationship you draw. Force connections between branches that didn't start together. Then test yourself with the map closed, because owning a beautiful diagram is not the same as being able to produce it.

The raw material for all of it is the set of passages you already thought were worth keeping. Glasp holds those highlights in one place across articles, PDFs, videos and Kindle books, so the map you build starts from evidence instead of memory. And because highlights on Glasp are public by default, you can see what other readers marked on the same source, which is the fastest way to find the link you missed.

Draw fewer maps. Label every line.