Working Memory Test — Free 2-Back Task Online

Soundary's N-back (2-back) test uses 8 letters, A to H. Each letter appears for 800 ms and disappears, and the next one follows 1700 ms later, so every letter takes 2500 ms and the full set of 20 letters takes under 1 minute. From the third letter on, the sequence is built so that each letter has about a 32% chance of being a match, the same as the letter two steps back, while non-matches are deliberately chosen to differ from that letter. The number of matches therefore varies from run to run, and the result screen shows how many there were this time.

There is a single Match button. You can press it once per letter, while the letter is on screen or during the blank gap before the next one, and it then shows that your press was registered. Pressing again on the same letter changes nothing, so press deliberately; when a letter is not a match, the correct response is to do nothing. If you switch tabs or apps, or tap Pause, the test stops; continuing discards the earlier responses and restarts with a new sequence, because the comparison with two letters back no longer holds once the chain is broken.

Accuracy, the headline score, adds your hits (pressing on a match) to your correct rejections (holding back on a non-match) and divides by 20. But the first two letters are never matches, and on average about 14 of the 20 letters are non-matches, so you can score fairly high without pressing at all. In a run with 6 matches, pressing nothing still gives 14/20, or 70%. Pressing on every letter, by contrast, only scores the share of matches. That is why accuracy alone can't tell you how hard your working memory actually worked.

The detailed report, unlocked by sharing or signing in, therefore shows four tiles (hits, misses, false alarms, correct rejections) and bars for the share of matches you caught (hit rate) and how often you pressed on a non-match (false alarm rate). As in signal detection theory, reading both rates together helps separate how readily you press from memory itself. Few hits with few false alarms suggests a cautious style that presses only when sure; many hits with many false alarms, a loose "press if it looks similar" style. The interpretation text changes at 90%, 75% and 60% accuracy: Soundary's own guide bands, while peer percentiles still await norm data.

The N-back is usually traced to Kirchner (1958), who used it to study age differences in holding rapidly changing information for a short time. It later became a staple of brain-imaging research on working memory: a meta-analysis of 24 studies by Owen et al. (2005) found that the task consistently engages dorsolateral and ventrolateral prefrontal cortex, premotor cortex and posterior parietal cortex, a network spanning frontal and parietal regions. That pattern comes from averages across many people, though, so one person's score is no basis for guessing the state of that person's brain.

Nor does the N-back measure the same thing as span tests, which ask how many items you can hold at once. Kane et al. (2007) found that 2-back and 3-back scores correlated only weakly with operation span, a standard complex span task, and a meta-analysis by Redick and Lindsey (2013) concluded that the two kinds of task can't be used interchangeably. Jaeggi et al. (2010) also noted that the N-back's reliability is too limited to rank individuals precisely. So read this result alongside span tests such as Digit span or Spatial span (Corsi), and across several runs rather than a single one.