The working-memory training task, free in your browser, with an honest account of what the research shows
Dual N-Back is the most researched — and most argued about — working-memory training task there is. In the full laboratory version two streams run at once, a square jumping around a grid and a spoken letter, and you flag whenever either one repeats what it was N turns ago. This version runs the position stream at 2-back: a square lights up in a 3 × 3 grid every beat, and you press Match when it is in the same cell as two beats earlier. Even one stream at 2-back means holding, updating and discarding on every beat, which is why it is exhausting after five minutes. New to n-back? Start with One-Back Letters or N-Color Back, and read the dual n-back guide for a turn-by-turn walk-through.
A square lights up in one of the nine cells of a 3 × 3 grid, one per beat.
Press Match (or the space bar) when the cell is the same one that lit up two beats earlier.
Do nothing on a non-match — silence is the correct answer most of the time.
Choose the block length and the pace before you start; a slower pace is the right place to begin.
At the end you see how many of the real matches you caught and how many false alarms you made.
A block is 20, 40 or 60 beats at a pace of 1.5, 1, 0.8 or 0.6 seconds per beat. On each beat after the second there is a 25% chance the square is deliberately placed in the cell from two beats back, plus the chance of a coincidental repeat.
Your percentage is hits divided by the real matches in the block. False alarms — pressing when it was not a match — are counted and shown separately but do not reduce the percentage, so a good run is a high percentage with few false alarms. Pressing on every beat will score well on the first number and badly on the second. Your best percentage is saved in this browser.
Working-memory updating: continuously replacing the contents of a limited store rather than just holding it. This is the mechanism the n-back task was built to load, and performance on the task itself improves substantially with practice. If you want the same process with a level that climbs automatically, play N-Color Back; for plain span rather than updating, try Digit Span. Another memory game to try next: Magic Hat.
The n-back task was introduced by Wayne Kirchner in 1958 to study short-term memory for rapidly changing information. The dual version, pairing a grid position with a spoken letter, became famous after a 2008 study by Jaeggi and colleagues reported that training on it raised scores on fluid-intelligence tests.
That claim has not held up well. A 2013 meta-analysis found reliable gains on the trained task and closely similar tasks, but no convincing transfer to general intelligence, and a 2016 review of the whole brain-training field reached the same conclusion. The honest summary: n-back is an excellent way to load working memory, you will get better at it, and it should not be sold as an IQ booster.
Do not try to remember a list. Hold a rolling window of exactly two cells and drop the oldest on every beat.
Say the positions under your breath — 'top-left, centre, top-left: match'. A verbal label is easier to hold than an image.
Accept that 2-back should feel hard. If it feels easy you are probably pattern-guessing rather than tracking.
Twenty minutes is the practical limit. Accuracy falls off after that and tired reps teach nothing.
Relaxed (1.5 seconds per beat). Start here with 20 beats. You have time to say each position in words, so build that habit now: label, compare with two back, drop the oldest. Move on when you catch nearly every match with no more than one or two false alarms.
Standard (1 second per beat). The labels have to get shorter — numbers 1 to 9 for the cells work better than "top-left". Try 40 beats to see whether your accuracy holds up as you tire.
Advanced (0.8 seconds per beat). There is no longer time to name and compare consciously, so rely on the rhythm. The usual failure is pressing on one-back repeats — the same cell twice in a row — which are not matches at 2-back.
Pro (0.6 seconds per beat). A beat every 0.6 seconds is close to as fast as the cell can be seen and judged. Use 60-beat blocks only when Advanced is solid, and judge yourself by false alarms as much as by the hit percentage.
This is the contested claim. An influential 2008 study reported gains on fluid-intelligence measures; later, larger and better-controlled work has largely failed to replicate them. You will reliably get better at n-back. Treat anything beyond that as unproven.
Most people see clear improvement on the task within one to two weeks of short daily sessions. That part is not controversial.
How many turns back you are comparing against. At 2-back you compare with two turns ago; at 3-back, three. Each step up is much harder than it sounds.
No — it is the position half, at a fixed 2-back. That is the honest place to start: the single-stream version is where almost everyone begins in studies, and it trains the same updating process.
Relaxed, at 1.5 seconds per beat, with a 20-beat block. Speed up one step at a time once you catch nearly all matches with very few false alarms.