What Is Working Memory, and Does N-Back Training Actually Work?

Working memory is the ability to hold information for a few seconds while working with it, like keeping a phone number in mind as you dial or carrying an intermediate result through a mental calculation. The short answer on n-back training is this: you get better at tasks similar to the one you trained, but the evidence that gains reach distant outcomes such as intelligence or school and work performance is weak. Below we look at what working memory is, how it is measured, and what the training research has shown.

Baddeley's model of working memory

The most widely used framework is the multicomponent model Baddeley and Hitch proposed in 1974. It has a phonological loop that keeps sound and language information alive through rehearsal, a visuospatial sketchpad that holds visual and spatial information, and a central executive that controls both stores and allocates attention. In 2000 an episodic buffer was added to bind information from several sources into a single scene.

The distinction matters because different tests tap different components. Repeating digits in order leans mostly on the phonological loop, tapping squares in the order they lit up on the visuospatial sketchpad, and judging whether a stimulus matches one from a few steps back in a constantly updating stream on the central executive. A single test score is therefore not working memory as a whole.

Capacity limits: seven plus or minus two, or four?

Miller's 1956 magical number seven, plus or minus two, is famous, but later research puts the number of independent items an adult can hold at once at about four, typically three to five, when rehearsal and chunking are prevented (Cowan, 2001). Digit spans come out around seven because rehearsing the digits as sound and grouping them into chunks are doing work.

Capacity grows through childhood, declines somewhat in old age, and within the same person shifts from day to day with sleep loss, anxiety, and fatigue. A trend over several runs therefore means more than any single score, and on a low-scoring day it is sensible to suspect your state before your ability.

Digit span, n-back, and spatial span: how they differ

They are all called working memory tests, but they ask for different things. Span tasks measure how much you can store, n-back measures how well you update a constantly incoming stream and suppress what is out of date, and the Sternberg task measures how quickly you search through what is in memory.

• 🔢Digit spanShort-term memory span for digits.Take the test →
• 🟦Spatial span (Corsi)Spatial working memory for locations.Take the test →

The training evidence: near versus far transfer

The n-back boom began in 2008 when Jaeggi and colleagues reported that a group trained for several weeks on dual n-back improved on a fluid intelligence test (Jaeggi and colleagues, 2008). Later large meta-analyses reached cautious conclusions. Melby-Lervag and Hulme (2013) found that working memory training produced short-term gains on verbal and visuospatial working memory tasks that did not last, and no convincing evidence of transfer to distant outcomes such as nonverbal intelligence, attention, reading, or arithmetic.

A meta-analysis restricted to n-back by Soveri and colleagues (2017) found the same pattern. Gains on tasks similar to the trained n-back were medium-sized, but effects on other working memory tasks and on fluid intelligence were small, and shrank further when compared against active control groups that had trained on a different task. In the researchers' own terms, it comes down to this.

• Near transfer, improvement on tasks that resemble the trained one: fairly consistent.
• Far transfer, improvement in distant areas such as intelligence, school, or work: small and unstable, and it shrinks as control groups get stricter.
• Expectancy effects: simply believing the training will help can nudge scores up, which is why active-control comparisons matter.

What actually affects working memory

More reliable than training is managing your state. Sleep loss is the best-known factor that lowers working memory and attention, and anxiety or worry consumes central executive resources and shrinks effective capacity. Regular aerobic exercise shows a small but consistent positive effect on executive function. Medications you are taking and alcohol can also affect a given day's score.

Strategy changes scores too. Chunking digits into groups of three or four, or giving them a rhythm, raises digit span noticeably. That is efficient packaging of information rather than a bigger store, so when you track your trend, keep the same strategy to make comparisons fair.

Tracking your trend on Soundary

The Soundary digit span test starts at 3 digits, adds one digit at a time up to 10, and ends after two misses at the same length; the result is the longest sequence you recalled. The spatial span test uses a 3-by-3 grid, runs from 3 to 9 squares, ends at the first miss, and reports the last length you completed.

The n-back test is a 2-back task with letters A through H, 20 trials long; responding when a letter matches the one two steps back counts as a hit, responding when it does not counts as a false alarm, and both are shown with your accuracy. The Sternberg test shows a memory set of 1 to 6 items and asks whether a probe was in the set, over 28 trials, then computes accuracy, mean reaction time, and the reaction-time slope per item. Repeating at the same time of day, on the same device, with the same strategy, a few weeks apart makes it much easier to separate day-to-day noise from real change.

• 🔁N-back (2-back)Working-memory updating.Take the test →
• 🧠Sternberg Memory ScanMemorize 1–6 digits, then instantly judge whether a probe digit was in the set. Beyond accuracy and reaction time, it measures how much each extra item slows your decision (ms/item) — revealing the speed and efficiency of your memory scanning.Take the test →

In short, working memory is a limited-capacity system that stores and manipulates at the same time, and each test taps a different part of it. N-back training makes you better at n-back, but the evidence that it raises intelligence is weak, so rather than betting on a training app, the realistic path is to look after sleep, exercise, and your general state while tracking your trend under matched conditions.