N-Back Test

The reference cue changes on every step of an n-back task. Each new position, letter or sound must be compared with the cue exactly N places earlier. At 2-back, the third cue is compared with the first, the fourth with the second, and so on. The new cue must also be remembered for a later comparison, so the task continually combines remembering, deciding and updating.

This use of working memory differs from learning a fixed list and recalling it afterward. The sequence keeps moving, and recognising something familiar is not enough to establish a match. Three terms distinguish the decisions:

Target

A cue that matches the one exactly N cues earlier. It calls for a match response.

Non-target

A cue that differs from its exact N-back reference. The correct action is to withhold a response.

Lure

A non-target that repeats a cue at a nearby, incorrect distance. Its familiarity can tempt a match response even though the required reference differs.

For example, consider the letters C, H, C, C in a 2-back task. The C at cue 3 matches the C at cue 1. The C at cue 4 must be compared with the second letter, H, so it is a non-target despite repeating the immediately preceding letter. Keeping the recent cues in order helps distinguish the required match from this kind of lure.

A single n-back task follows one stream, such as locations. A dual task requires two independent decisions, such as location and letter identity. A repeated location can contain a different letter, and a repeated letter can appear somewhere else. Either stream, both streams or neither stream may match. Treating the location and letter as one inseparable pair would miss valid matches in just one stream.

The conditions change what the task asks of the participant. Increasing N means keeping track of a more distant reference. A shorter interval allows less time to compare and update, while adding a second stream requires another decision at every step. These differences make n-back useful for demonstrating continuous memory updating in education and cognitive experiments, but they also prevent percentages from different conditions being treated as equivalent.

Errors can involve overlooking a real match or responding to a non-target. Overall accuracy includes correct decisions to remain silent, so it can conceal poor detection of targets when non-targets are common. Results describe performance on the particular task and conditions. A brief practice result is not a clinical assessment, an intelligence score or a complete measure of working-memory ability, and a higher score alone does not establish improvement in everyday abilities.

How to Use This Tool:

Choose whether to follow positions alone or make separate position and letter decisions. Both modes are visual. Dual mode uses displayed letters with no spoken audio, making it a visual adaptation rather than an audiovisual research protocol.

1. Choose Single · position or Dual · position + letter under Streams. Positions occupy the eight outer squares of a 3 × 3 grid; the centre shows session progress.
2. Set Compare how far back to 1 through 5 and choose 20 or 40 Scored cues. The defaults are single mode, 2-back and 20 scored cues. Every session also begins with N memory-only cues.
3. Open Advanced if you want to change Cue interval from its 3000 ms default. The allowed range is 1500 to 5000 ms, with whole-millisecond entry or a slider in 100 ms steps.
   If the summary says Check setup, correct the settings before starting. For example, a 1000 ms interval is outside the allowed range; entering 3000 ms restores a valid interval.
4. Select Start session when ready and keep the tab visible and the window focused. Settings remain fixed during the session. Remember the first N cues without responding; the response buttons are disabled until comparisons can begin.
5. Press A or the Position button for a position match. In dual mode, press L or Letter for a letter match. Respond to both if both match, and do nothing for a non-target. A cue is shown for 500 ms, but responses remain possible during the masked remainder of its interval. The first accepted response in each stream is final.
6. Wait for completion or select Stop to end early. Stop does not pause the session. When percentages appear, check Stream decisions for missed targets and false alarms; a Partial session summary means the planned session was not completed.

Interpreting Results:

Read the position and letter results separately. In dual mode, the summary lists position accuracy first and letter accuracy second; these are not combined into one score. Hit rate shows how many actual targets received a response. False alarm rate shows how often non-targets received an incorrect response.

A completed generated session contains 30% targets per stream. Withholding every response therefore produces 70.0% accuracy while missing every target, with a hit rate of 0.0%. Check the hit rate and false alarm rate together before interpreting the headline percentage. If misses predominate, inspect the missed targets and their references. If false alarms predominate, check whether responses followed a repeat at the wrong distance.

Trial evidence shows the current cue and its exact N-back reference alongside target and lure labels, the response, outcome, input method, latency and scoring inclusion. Its CSV and DOCX exports retain that context for checking individual decisions. The table CSV exports retain unrounded numeric values, so their precision can exceed the one-decimal display.

Error path accumulates misses and false alarms separately for each stream. A rise means another error; a flat section means no additional errors. The curve cannot fall and does not by itself establish fatigue or improvement. Its image export shows the progression, while its CSV contains the per-cue cumulative counts.

For repeat comparisons, keep mode, N, session length and cue interval consistent, and compare completed sessions separately from partial ones. Each start generates a fresh schedule, so matching settings do not reproduce the same cue order. In a completed 20-cue stream, changing one decision changes accuracy by 5 percentage points, and changing the response to one of its six targets changes hit rate by about 16.7 points. Small changes in counts can therefore produce noticeable percentage changes.

Technical Details:

Each completed, eligible cue receives one outcome per active stream. The outcome depends on the exact N-back comparison and whether a match response was accepted within that cue's response window.

Formula Core

For one stream, H denotes hits, M misses, F false alarms and C correct rejections. S is the number of eligible scored cues. Every such cue contributes to exactly one of the four counts:

Accuracy is the percentage of eligible cues answered correctly, including correct withholding:

Hit rate uses only targets as its denominator. It measures the share of targets detected:

False alarm rate uses only non-targets as its denominator. It is the share of non-targets incorrectly marked as matches, not the share of match responses that were wrong:

Counts are whole numbers, and available rates range from 0% to 100%. Percentages are calculated before rounding and displayed to one decimal place. A zero denominator makes the corresponding rate unavailable, shown as an em dash. No eligible cues means no accuracy; no targets means no hit rate; no non-targets means no false alarm rate.

Target scheduling and lures

A generated session has exactly six targets among 20 scored cues, or 12 among 40, in each active stream. Target cue positions are selected without replacement. Target values repeat their exact N-back reference, and non-target values are chosen to differ from it, preventing additional accidental targets. Position and letter target schedules are selected separately, so simultaneous matches are possible but not required.

The position choices are the eight outer grid cells. In Trial evidence, positions are numbered left to right and top to bottom from 1 to 9, with the centre, 5, unused. The letter choices are C, H, K, L, Q, R, S and T. A position match compares cell location regardless of letter, and a letter match compares letter identity regardless of location.

A lure repeats a cue at N − 1 or N + 1 places back while differing from the cue exactly N places back. The earlier cue must exist, and zero-distance comparisons are excluded. Thus, at 1-back, a two-back repeat can be a lure. Lures occur incidentally and have no fixed quota. They remain non-targets and receive the ordinary non-target scoring rule.

The 30% target proportion applies to a completed generated session. An early ending can leave a different proportion among the scored cues, and the labelled worked example is a separate short demonstration.

Response windows and exclusions

The intended display time is 500 ms per cue. The selected interval covers both that display and the masked period afterward. At the default 3000 ms interval, 2500 ms remains after masking. Responses are accepted from the recorded cue onset up to, but not including, the interval deadline. A response at or after that deadline is ignored even if the next cue has been delayed.

The first N cues provide the initial history and are wholly unscored. Afterward, each stream accepts at most one response per cue. Further presses cannot add another response or undo the first, and held-key repeats are ignored. Withholding requires no button press and becomes a miss or correct rejection when the cue closes.

Selecting Stop, hiding the tab or losing window focus ends the session. A cue-ending timer running more than 250 ms late also causes an interruption. The active unfinished cue is excluded in full, even if a response was already recorded. Earlier closed scored cues remain in the partial result. The unfinished cue appears as Interrupted with Scored set to No; it does not become a miss. There is no automatic resume.

Latency (ms) is the browser's elapsed time from its recorded cue onset to the first accepted response, including responses that become false alarms. No response has no latency value. Displayed latencies use one decimal place and do not weight the score. Browser scheduling, display and input delays are uncalibrated, so these are browser observations rather than calibrated measurements of physical reaction time.

Privacy Notes:

Cue generation and scoring take place in your browser. The task does not upload responses or results. Observations remain in the current tab and clear on reload; starting a new session or loading the worked example replaces the current observations. There is no saved personal session history.

Only mode, N, scored-cue count and interval enter the URL. Sharing it shares the setup, not the cue order, responses or results. Download evidence before replacing a session if you want to retain it. Downloaded documents and copied data are separate copies that you control.

Worked Examples:

References:

• Working memory, attention control, and the N-back task: a question of construct validity, American Psychological Association, May 2007.
• N-Back task / 2-Back task, PsyToolkit, July 16, 2026.
• What is Dual N-Back?, XNBack.