Dual N-Back is a working-memory task where you track two independent streams of stimuli (usually a letter and a spatial position) and press a button whenever the current item matches the one that appeared N steps back.

Does Dual N-Back improve intelligence?

A 2008 study by Jaeggi et al. suggested it might, but later replication attempts have largely failed to reproduce that finding. The current evidence supports that practice improves performance on the N-back task itself, with limited transfer to other cognitive tasks.

Dual N-Back — Free Working Memory Trainer

Track whether the current letter and position match N steps back. The most-studied working-memory task. A letter is spoken and shown in one of 9 positions on each trial. Press A (or the Audio button) if the letter matches the one from N steps back. Press P (or the Position button) if the position matches.

What Dual N-Back measures

Dual N-Back is a working-memory task that requires you to continuously update and monitor information in real time. On each trial (typically presented every 3 seconds), two things happen simultaneously: a letter appears in one of nine positions on a 3×3 grid, and the same letter is spoken aloud. Your job is to track whether the current letter matches the one that appeared N steps back, and whether the current position matches the one from N steps back. You respond with two independent buttons.

The "dual" part is what makes this hard. Doing either task alone is moderately demanding; doing both at once requires dividing attention across two simultaneous streams while continuously updating a sliding window of the last N items. This is fundamentally different from digit-span or Corsi-span tasks, which ask you to store a static sequence — N-Back asks you to update stored information against an ongoing stream.

The 2008 Jaeggi study and the debate that followed

In 2008, Susanne Jaeggi and colleagues published a study suggesting that practicing dual N-Back for several weeks produced gains in fluid intelligence — the ability to solve novel problems, independent of acquired knowledge. This was a striking claim because fluid intelligence had long been considered largely fixed in adulthood. The study received enormous press coverage and spawned an entire brain-training sub-industry around N-Back apps.

Subsequent replication attempts have been largely disappointing. A 2013 meta-analysis (Melby-Lervåg & Hulme) and a 2018 cross-sectional study (Sala et al., PMC) found that N-Back practice improves performance on N-Back tasks but does not reliably transfer to fluid intelligence or other working-memory measures. The original Jaeggi finding has not been decisively refuted, but the burden of evidence has shifted toward skepticism. We tell you this honestly because the niche is full of apps that still cite the 2008 study as settled science; it isn't.

Worked example (N=2)

Suppose the sequence of (letter, position) is: (C, top-left), (F, center), (C, bottom-right), (H, center), (F, bottom-right). On trial 3, the letter C matches trial 1 (2 back) — so you press Audio. On trial 5, the position "bottom-right" matches trial 3 (2 back) — so you press Position. Most trials will have no match on either stream.

How to approach the task

Start at N=2. N=1 is too easy and gives no real practice; N=3 is hard for first-timers. N=2 is the standard starting point.
Don't try to remember the whole sequence. You only need the item N back. Rehearse just that one and let the older ones go.
It's fine to miss some. Even trained participants miss 20-30% of matches. Accuracy below 70% usually means N is too high for you.
Use both modalities. Read the letter silently and note the position visually — they're independent streams.

Adjusting difficulty

The conventional rule (from Jaeggi's protocol) is: if you score above 80% accuracy on a session, increase N by 1 next session. If you score below 50%, decrease N by 1. This keeps the task at the edge of your ability, which is where learning is most efficient.

Realistic expectations Dual N-Back is genuinely demanding and feels productive. But the strong intelligence claims made for it in 2008 have not held up well. If you enjoy it, the engagement is real and the task does exercise working memory. Just don't expect IQ gains — and don't trust any app that promises them.