Explainer
Dual N-Back: How It Works, What the Research Shows, and How to Start
By Alexander Gusev, founder of IQMogPublished Updated
Dual n-back is a working-memory game: squares light up on a grid, letters are spoken, and you tap when either matches the one from N turns back. Across 33 trials, training gave a medium-sized gain on untrained n-back tasks. Gains on other memory tasks and on IQ tests were very small.
| Fact | Value | Source |
|---|---|---|
| The study that started the craze | 70 adults recruited, 69 analyzed; about 25 minutes a day for 8 to 19 days; reasoning gains grew with training | Jaeggi et al., PNAS, 2008 |
| n-back meta-analysis | 20 studies, 559 trained and 463 control participants; reasoning effect g = 0.24 | Au et al., Psychonomic Bulletin & Review, 2015 |
| Working-memory training meta-analysis | 87 publications, 145 comparisons; gains on verbal and visuospatial memory tests, reasoning g = 0.05 against treated controls | Melby-Lervåg, Redick and Hulme, Perspectives on Psychological Science, 2016 |
| n-back transfer meta-analysis | 33 trials, 203 effect sizes; medium effect on untrained n-back tasks, very small on other tasks | Soveri et al., Psychonomic Bulletin & Review, 2017 |
| Reasoning gain against active control groups | g = 0.06, an IQ-scale equivalent under 1 point (0.06 × 15 = 0.9) | Our arithmetic from Au et al., 2015 |
| Free dual n-back programs | Brain Workshop, free and open source for Windows, Mac OS X and Linux | Brain Workshop, undated |
What is dual n-back?
Dual n-back is a working-memory game: a grid position and a spoken letter play together, and you flag matches from N steps back.
Au and colleagues describe working memory, the workspace where you hold and update information, as an important predictor of scholastic aptitude and a critical bottleneck for higher-order processes such as controlled attention and reasoning. The n-back task keeps that workspace busy: you hold the last N items and compare each new one against them. In the 2008 study that made the game famous, a square appeared at one of eight screen locations while a consonant played through headphones, one new pair every 3 seconds. “Dual” means two streams at once. Single n-back uses one.
A 2-back example in five steps
In this made-up run, N = 2, so each turn is compared with the one two before it.
- Turn 1: the square lights the top-left cell and you hear “K”. Nothing to compare yet.
- Turn 2: the square lights the center cell and you hear “T”. Still nothing two turns back.
- Turn 3: the square lights the top-left cell and you hear “R”. The position matches turn 1, so you tap the position button.
- Turn 4: the square lights the bottom-right cell and you hear “T”. The letter matches turn 2, so you tap the letter button.
- Turn 5: the square lights the top-left cell and you hear “R”. Both match turn 3, so you tap both buttons.
For how working memory sits next to reasoning, see what fluid intelligence is.
How do you play dual n-back?
You play dual n-back by tapping a position or letter button whenever the current item matches the one from N turns back.
Every n-back test follows the same loop, and the research versions add two features that matter for you: adaptive levels and long sessions made of many short blocks. In the 2008 study, the program adjusted N to each person’s performance, and one session was 20 blocks, about 25 minutes a day.
- Start at 1-back or 2-back. At 1-back you compare each turn with the one just before it. At 2-back you compare it with the one before that.
- Tap only on real matches. A position match and a letter match are separate answers, and a turn can have both, one or neither.
- Let the level follow your play. When you are accurate, N goes up. When you slip, it comes back down, so you always train near the edge of what you can hold.
- Finish the block, then look at your result before starting the next one.
A guided tutorial helps in the first session, because the double stream is easier to feel than to read about.
Does dual n-back work?
Dual n-back improves n-back performance, carries over modestly to other memory tests and moves reasoning-test scores only slightly.
| Study and who took part | Gain on the trained task | Gain on untrained n-back tasks | Gain on other working-memory tasks | Gain on reasoning tests |
|---|---|---|---|---|
| Jaeggi et al. 2008: 70 healthy adults recruited, 69 analyzed (34 trained, 35 untrained controls) | All four training groups improved on the dual n-back task | Not tested | Digit span: trained groups gained more than controls (P < 0.001). Reading span: both groups improved equally | Trained groups gained more than controls: significant after 17 and 19 days (P < 0.05), a trend after 12 days (P = 0.09). Controls improved too |
| Au et al. 2015: 20 n-back studies, 559 trained and 463 control participants | Not the outcome (studies had to report a reasoning test) | Not the outcome | Not the outcome | g = 0.24 overall, about 3 to 4 IQ points. g = 0.06 with active controls, 0.44 with passive controls |
| Melby-Lervåg, Redick and Hulme 2016: 87 publications, 145 comparisons, all working-memory programs | Not isolated for n-back | Not isolated for n-back | Verbal working memory g = 0.31 (60 comparisons) and visuospatial g = 0.28 (40) against treated controls | Nonverbal ability g = 0.05 against treated controls, 0.20 against untreated controls |
| Soveri et al. 2017: 33 randomized trials, 203 effect sizes, healthy adults | Transfer effects only | Medium-sized effect | Very small effect | Very small effect. Age, training dose, training type and task content did not change the result |
The 2008 study is where the idea began. Gains on the matrix reasoning tests rose with the number of training days, and the untrained control group improved as well, which the authors put down mainly to retest effects.
Can dual n-back increase IQ?
Dual n-back does not reliably increase IQ: against active control groups, the meta-analyses find reasoning-test gains of about 2 IQ-scale points at most.
Effect sizes are in standard-deviation units, and the IQ scale uses 15 points per standard deviation, so g × 15 gives the IQ-scale equivalent. The table shows how the headline number shrinks once the comparison group does something.
| Result | Effect size (g) | IQ-scale equivalent (g × 15) |
|---|---|---|
| Au 2015, all 20 studies | 0.24 | 3.6 (the paper says 3 to 4) |
| Au 2015, studies with passive controls | 0.44 | 6.6 |
| Au 2015, studies with active controls | 0.06 | 0.9 |
| Melby-Lervåg 2016, all programs against untreated controls | 0.20 | 3.0 |
| Melby-Lervåg 2016, all programs against treated controls | 0.05 | 0.75 |
| Melby-Lervåg 2016, n-back studies against treated controls | 0.15 | 2.25 (1.5 at g = 0.10 after removing its most problematic study) |
A group that does nothing shows what happens with no extra effort. A group that practices a different task shows what n-back adds. Au’s team found the passive-control advantage tracked how much participants were paid and where the study ran: adjusting for both cut it from 0.31 to 0.09, no longer significant. Melby-Lervåg and colleagues found no convincing far-transfer gains against treated controls. For what moves a score on a test, see what moves an IQ test score; practice effects versus real gains splits the two.
What you get is measured: a medium-sized gain on untrained n-back tasks (Soveri et al., 2017) and gains on verbal and visuospatial working-memory tests straight after training (Melby-Lervåg et al., 2016). If you want a number for your reasoning today, the 20-puzzle test is the next step.
How long should you train?
Plan about 20 to 25 minutes a day on weekdays for three to four weeks, the rhythm of the original study’s longer courses.
The 2008 study trained people for about 25 minutes a day on weekdays, for 8, 12, 17 or 19 days. Nineteen weekdays is just under four weeks (19 ÷ 5 = 3.8). Au’s team looked at what predicted the reasoning effect across 20 studies, and neither session length nor the number of sessions did so significantly (Au et al., 2015). The shortest session in their sample was 18.5 minutes, so the data say nothing about shorter ones.
- Block length: let one block be a short, complete run, and stack blocks until you reach 20 to 25 minutes.
- Frequency: train on weekdays, the schedule the original study used.
- Level: stay where you are right most of the time and still have to work. Let the adaptive rule move N.
- Length of the plan: a four-week course gives you a clean before and after on your top level.
The studies in Au’s analysis recruited healthy adults aged 18 to 50, with a mean age of 22.85. For how mental abilities shift across later decades, see how IQ changes across adulthood.
Which dual n-back app should you use?
A good dual n-back app adapts to your level, runs on your phone and makes daily training easy; the IQ Mog app does all three.
Brain Workshop is the long-standing free option. Its own page calls it a free, open-source dual n-back program for Windows, Mac OS X and Linux, first released in August 2008. If you want to train at a desk, it is a solid place to start. If you want to train on your phone, the IQ Mog dual n-back app gives you positions on a grid plus spoken letters, a guided tutorial, and an automatic mode that adjusts difficulty as your performance changes.
| Option | Runs on | Cost | Training features |
|---|---|---|---|
| Brain Workshop | Windows, Mac OS X and Linux | Free and open source | A full dual n-back program |
| IQ Mog | iPhone and iPad (iOS or iPadOS 16.4 or later) and Android | Tutorial and first two complete training blocks free. IQ Mog Pro unlocks unlimited training, weekly or yearly, with prices shown in the store | Guided tutorial, automatic difficulty, daily goal, optional reminder and training history. No account; scores and history stay on your device; analytics are off by default |
The routine from the last section fits either one. Open the app, run your blocks, watch your level climb, and keep the habit going with a daily goal. After four weeks, take the 20-puzzle test and see where your reasoning sits today.
Frequently asked questions
What is a good n-back level?
There is no fixed benchmark in the studies. In the 2008 experiment, the program adjusted N to each person’s performance, so a good level is the one where you are right most of the time and still have to work for it. Start at 1-back or 2-back and let accuracy move you up.
Is dual n-back hard?
Yes, by design. The 2008 study describes it as a demanding working-memory task, and adaptive levels exist so that every player trains near their own limit. A guided tutorial and a start at 1-back make the first sessions manageable.
How many minutes a day should I train?
The original study used about 25 minutes on weekdays (Jaeggi et al., 2008), and the shortest session in the 20 studies of Au et al. was 18.5 minutes. About 20 to 25 minutes on most days fits that range.
Is single n-back as good as dual n-back?
The reviews found no significant difference. Au and colleagues found none between studies using dual and single n-back for reasoning transfer, and Soveri and colleagues found that training type did not moderate their results either.
Does dual n-back make you better at everyday memory?
It makes you better at the memory tasks closest to the game. Melby-Lervåg and colleagues found reliable gains on verbal and visuospatial working-memory tests straight after training, and the largest effect in Soveri’s analysis was on untrained n-back tasks.
Sources
- Proceedings of the National Academy of Sciences 105(19), 6829-6833 (via PubMed Central): Improving fluid intelligence with training on working memory (2008-05-13; accessed )
- Psychonomic Bulletin & Review 22(2), 366-377: Improving fluid intelligence with training on working memory: a meta-analysis (2015-04; accessed )
- Perspectives on Psychological Science 11(4), 512-534 (via PubMed Central): Working memory training does not improve performance on measures of intelligence or other measures of far transfer: evidence from a meta-analytic review (2016; accessed )
- Psychonomic Bulletin & Review 24(4), 1077-1096: Working memory training revisited: a multi-level meta-analysis of n-back training studies (2017; accessed )
- Brain Workshop (SourceForge): Brain Workshop: a dual n-back game (undated; accessed )
- Brain Workshop (SourceForge): Brain Workshop release news (2008-08-09; accessed )
Next step
Take the IQMog test
IQMog gives you an IQ estimate with a score range from 20 Raven-style visual pattern puzzles. Take it free in about 10 minutes, then get your full report with percentile and score range for $4.99, once. No subscription.
Related reading
IQ research
- IQ score by countryExplore national IQ estimates for 249 countries and territories, with the data behind each one.
- What IQ scores predictThe research on IQ and school, work, income and health, with the size of each link.
- How to Get Your Best Online IQ Test Result (Without Fake Hacks)A practical system for a cleaner online IQ test result with better setup, pacing, practice on the format, and a repeatable pattern-solving framework.