Does dual n-back training work? What the evidence actually says

The question behind the question

When people ask whether dual n-back works, they rarely mean “will I get better at dual n-back”. They mean something larger: will my memory improve, will I think faster, will I be smarter. Those are different questions, and the research answers them very differently. Getting better at the exercise itself is close to certain if you practise it. Whether that improvement travels outward to anything else is one of the most contested questions in cognitive psychology, and it has been argued in print for nearly two decades. This guide separates the two questions, because almost every confused claim about n-back training comes from letting them blur together.

Where the excitement started

In 2008, Susanne Jaeggi and colleagues published a study in PNAS in which healthy young adults trained on a dual n-back task for between eight and nineteen sessions. The trained groups improved on a matrix-reasoning test more than untrained controls, and the improvement appeared to scale with the amount of training. The paper was striking because fluid reasoning had long been treated as largely fixed in adults. It was widely covered, and it is the origin of most of what you will read online about n-back training today. Understanding what happened next matters more than the original result.

What the task is actually made of

A dual n-back trial gives you two things at once: a square lights up somewhere on a three-by-three grid, and a letter is spoken. Your job on each turn is to answer two independent questions. Is this position the same as the one N turns ago? Is this sound the same as the one N turns ago? Nothing is stored for you. To answer, you must hold a rolling window of the last N positions and the last N sounds, add the newest pair, discard the oldest, and keep the order intact while a new pair arrives every couple of seconds. That constant rewriting is the reason the task is tiring in a way that memorising a list is not. It is also why the research literature classes it as an updating task rather than a storage task.

Why the task is hard to game

Most memory games can be beaten with a trick. You can chunk digits, build a story, or place items in a familiar room. Dual n-back resists this because the window keeps moving. A mnemonic that works for a fixed list becomes a liability when the list is rewritten every two seconds, since the effort of maintaining it competes with the updating itself. Kane and colleagues examined how n-back performance relates to attention control in 2007 and found the relationship is real but not simple, which is part of why the task is used in research at all. In practice, most people who improve report that they stopped using an explicit strategy rather than that they found a better one.

What transfer means, and why it decides everything

Researchers describe improvement in two ways. Near transfer means the gain shows up on tasks that closely resemble the trained one: another n-back variant, a similar updating task. Far transfer means the gain appears on something genuinely different, such as a reasoning test, reading comprehension or attention in daily life. Near transfer is comparatively easy to produce and easy to explain, because you are partly measuring familiarity with the task format. Far transfer is the claim that makes brain training commercially interesting, and it is the claim that has failed most often when tested carefully.

Near transfer: the part nearly everyone accepts

Practise an n-back task and your n-back scores rise. Practise it enough and you improve on tasks that share its structure. This is not controversial, and it is not trivial either: it means the task is learnable and that your performance is a real measurement of something you are doing better. What it does not establish is the cause. You may be improving your ability to hold and update items, or you may be developing a strategy, becoming faster at the interface, or simply learning what the task feels like when you are about to make an error.

Far transfer: where the evidence thins out

In 2013, Redick and colleagues ran a larger study with an active control group, meaning the comparison group also did a demanding computer task rather than nothing at all. They found no improvement in fluid intelligence after n-back training. The same year, Melby-Lervåg and Hulme published a meta-analysis of working-memory training across many studies and reported short-lived near effects and no reliable far transfer. Later work by the same authors in 2016 reached a similar conclusion after pooling a larger set of studies and paying close attention to control-group design.

Why the meta-analyses disagree

In 2014, Au and colleagues pooled n-back studies specifically and reported a small positive effect on fluid intelligence. In 2016, Melby-Lervåg, Redick and Hulme pooled a broader set and reported essentially none. In 2017, Soveri and colleagues published a multi-level meta-analysis of n-back training and found clear improvement on n-back tasks themselves, some improvement on closely related tasks, and little beyond that. The disagreement is not mysterious. It comes from which studies are included, whether control groups were active or passive, and how outcomes are grouped. Small differences in those decisions move the headline number a lot.

The control group problem

If your comparison group does nothing while the training group turns up for twenty sessions, any difference at the end can come from expectation, motivation or simply being tested twice under different conditions. Active control groups exist to remove that explanation. As a rule of thumb across this literature, the more carefully the control group is matched, the smaller the reported far transfer becomes. That pattern is itself informative: it suggests a large part of the early enthusiasm was measuring the effect of taking part in a training study rather than the effect of the training.

What the field concluded

In 2016, a group of researchers led by Daniel Simons published a long review in Psychological Science in the Public Interest examining brain-training claims as a whole. Their conclusion was that the evidence supports improvement on trained tasks, provides some support for closely related tasks, and does not support broad claims about improved cognitive performance in everyday life. That review is worth reading in full if you want the full picture rather than a headline, and it applies to commercial brain-training suites as much as to n-back specifically.

So is it worth doing?

That depends on what you want from it. If you want a demanding, measurable exercise that gives you an honest number each session and gets harder as you improve, dual n-back does that well and does it for free. If you want a proven route to better reasoning, better grades or a sharper mind at work, the evidence does not currently support that promise, and you should be suspicious of any app that makes it. Practising something difficult and tracking your accuracy is a reasonable thing to do. Expecting it to change your life is not.

What people report anecdotally, and how to read it

Search for dual n-back and you will find people describing sharper focus, easier reading, better conversation recall. These reports are sincere and they are also exactly what you would expect from any regular, demanding, self-chosen practice that you believe is working. Sleep, exercise and simply paying more attention to your own attention produce the same impressions. None of that makes the reports worthless, but it does mean they cannot settle the question. The only way to separate a real effect from an expectation effect is a controlled comparison, which is precisely what the studies above tried to do and why their results carry more weight than a forum thread.

What would change the picture

This is not a closed question, and it would be wrong to present it as one. Several things could reasonably shift the consensus: larger pre-registered trials with active controls, longer training periods than the few weeks most studies use, better matching of participants, and outcome measures chosen before the data is collected rather than after. Individual differences may also matter more than the averages suggest, since a pooled result of roughly zero can hide people who improved and people who did not. Until that work exists, the responsible summary is the one the reviews give: reliable near effects, unreliable far effects.

A reasonable way to think about it

Treat dual n-back the way you would treat a difficult crossword, a musical instrument or a sport you are not going to compete in. It is worth doing if the doing is worth something to you: the concentration it demands, the visible progress, the small satisfaction of a clean session at a level that used to be impossible. Those are real returns and you do not need a research paper to claim them. What you should not do is buy an app, or defend one, on the promise of a smarter brain. The exercise is honest. The marketing around this whole category often is not.

Training without fooling yourself

Three habits keep the exercise honest. First, judge yourself on accuracy at a fixed level rather than on the highest level you have ever reached, because level alone rewards guessing. Second, compare sessions done under similar conditions, since sleep, caffeine and time of day move your scores more than most people expect. Third, treat any improvement you notice elsewhere as unproven until you have a reason to believe it. Dual N-Back shows your accuracy per session so that you can follow the exercise itself, which is the part the evidence actually supports.

Further reading

• Jaeggi, Buschkuehl, Jonides & Perrig (2008), Improving fluid intelligence with training on working memory, PNAS
• Redick et al. (2013), No evidence of intelligence improvement after working memory training, Journal of Experimental Psychology: General
• Melby-Lervåg & Hulme (2013), Is working memory training effective? A meta-analytic review, Developmental Psychology
• Au et al. (2015), Improving fluid intelligence with training on working memory: a meta-analysis, Psychonomic Bulletin & Review
• Melby-Lervåg, Redick & Hulme (2016), Working memory training does not improve performance on measures of intelligence or other far transfer, Perspectives on Psychological Science
• Soveri et al. (2017), Working memory training revisited: a multi-level meta-analysis of n-back training studies, Psychonomic Bulletin & Review
• Simons et al. (2016), Do “Brain-Training” Programs Work?, Psychological Science in the Public Interest