What the test measures
Letters and digits flash one after another in the same place, ten a second. Two of the items are letters, and at the end you report both. When the second letter comes 200 to 500 ms after the first, people are much less likely to report it, although at other times they see it easily. Raymond, Shapiro and Arnell (1992) called this gap the attentional blink. This page measures how deep your blink is and how long it lasts.
How a trial runs
- A cross appears in the centre for 500 ms. Keep your eyes on it.
- 19 items follow in its place without gaps, each for 100 ms: digits from 2 to 9, never the same digit twice in a row, and two different consonants, the targets. The consonants come from B C D F G H J K L M N P R S T X Z: no vowels and no Q, V or Y.
- The first letter is item 5, 6, 7 or 8. The second comes 1, 2, 3, 5 or 8 items later (the lag), so 100 to 800 ms after it, and at least three digits follow it.
- Then you pick the two letters. Their order does not count. Guess when you are not sure.
Standard has 15 trials for each lag, 75 in all, in random order with a break half way, about 9 minutes. Short has 9 for each lag, 45 trials. Before your first run there are 8 practice trials with feedback, the first 4 slower (150 ms an item); they do not count.
In the experiment of Raymond, Shapiro and Arnell (1992), black letters flashed on a grey field for 15 ms each with 75 ms of grey between them, 11.1 letters a second. The first target was the only white letter; the second was an X that people only had to detect. When they also had to name the white letter, they detected the X on fewer than 60% of trials when it came 180 to 450 ms later. When told to ignore the white letter, they detected the X on 85% or more of trials at every position. Here both targets are letters among digits, and each item stays until the next one replaces it.
Every frame is checked
A screen draws whole frames, so an item lasts a whole number of them: 6 frames at 60 Hz, 5 at 50 Hz, 8 at 75 Hz (106.7 ms) and 14 at 144 Hz (97.2 ms). Before each run the page measures the frame rate over at least 120 frames, then records the frame in which every item was drawn. If any item from the one before the first letter to the second one after the second letter came late, stayed a different number of frames or had a frame lost while it was shown, the trial does not count and is repeated once later: such an item changes the time between the letters, which is what the test measures. Above 10% mistimed trials the result warns you; above a quarter, the run ends without a result.
How the blink is measured
For each lag the page takes the trials in which you reported the first letter and counts how often you also reported the second: the share T2|T1. Trials with the first letter missed are left out, because then there was nothing to blink from. The blink size is how far T2|T1 at lags 2 and 3 falls below your accuracy on the first letter over all trials, T̄1, as a share of T̄1, averaged over the two lags:
An example: the first letter right in 90% of trials and the second in 45% at lag 2 and 60% at lag 3 give ½ × (0.45 / 0.9 + 0.30 / 0.9) × 100% = 41.7%. A blink size near 0 means no dip. Lag-1 sparing is T2|T1 at lag 1 minus the mean of lags 2 and 3, in percentage points.
The 95% interval of each T2|T1 is Wilson's interval. For the blink size the page draws the three shares 2,000 times from beta distributions fitted to your counts (Jeffreys' prior: half a trial added to each side) and keeps the middle 95% of the results. In simulated runs with 15 trials per lag (four kinds of participant, 1,000 runs each, 1 November 2026) this interval held the true blink size in 94 to 98% of runs, and it was 20 to 45 percentage points wide.
What to look for
- Dux and Marois (2009) describe the blink as lower T2|T1 at lags 2 to 5 than at lags 6 to 8. The shaded band in the chart marks 200 to 500 ms after the first letter.
- Lag 1: Raymond et al. already saw that a second target right after the first was hardly affected. Dux and Marois call this lag-1 sparing.
- The dashed line is your accuracy on the first letter; the dip is measured against it.
Why it happens
Dux and Marois (2009) conclude that the first target makes demands on attention while it is selected, stored in working memory, registered as an event and prepared for the answer. These steps share one limited central resource, so a second target arriving 200 to 500 ms later has to wait and can be overwritten by the digits that follow. In one model they review, a second target straight after the first gets the same boost as the first, which explains lag-1 sparing.
Brief display: Sperling's partial report
The third task asks how much you see in one glance. A grid of 12 consonants, 3 rows of 4, flashes for 50 ms (3 frames at 60 Hz). In the first part you type every letter you saw, in any order: the whole report, 8 trials. In the second an arrow points to one row and you type its 4 letters from left to right, guessing those you are not sure of; a letter counts only in its own place. The arrow comes 50 ms before the letters, or 0, 150, 300, 500 or 1,000 ms after they vanish, 6 trials each in random order, after 6 practice trials with the arrow first.
You do not know which row will be asked for, so the share right in that row samples the whole grid: available letters = share right × 12. Sperling (1960) used tones of 2,500, 650 and 250 Hz instead of an arrow, grids of letters and digits, a tachistoscope, and five observers (four students and a member of the faculty) over 12 sessions. In whole reports people wrote about 4.5 symbols. With the tone 50 ms before a 12-symbol display, 82% of the symbols were right, 9.8 available; with the tone 1 s after it, 38%, close to a whole report. The chart marks these values. The gap between an early and a late arrow is a brief visual memory that fades within about a second.
Things to try
- Switch to Second letter only (30 trials at lags 2, 3 and 8) and report only the second letter. In Raymond et al.'s control, people who could ignore the first target showed no dip. Here you still have to notice the first letter to know which one is second, so compare the dip with your full run.
- Run the test on two different days on the same screen and compare the intervals in History. When they overlap, the difference may be chance.
- In the brief display, compare your whole report with the letters available at 0 ms: the gap is what you saw but could not hold long enough to report.
Comparing your result
History compares runs of the same task on the same input whose refresh rates are within 10% (or both below 50 Hz) and whose items lasted the same number of frames. There is no percentile or score against other people: published studies used other screens, rates, letters and numbers of trials, and a run of 75 trials is too short for a ranking. Compare your own runs and their intervals.
Limits
- Timing depends on your screen. Below 50 Hz, or with uneven frames, the page offers Run anyway: items then last whole slower frames (3 frames = 92 ms at 32.6 Hz), and the run is marked less precise and kept apart in History. Below 25 Hz the test does not run, because an item would last 2 or 3 frames and one lost frame would change it by a third or more. Remote desktop may merge or skip brief frames on your screen; the page cannot detect it.
- Liquid-crystal screens need some milliseconds to change from one item to the next, so a letter may leave a faint trace; the page cannot measure this.
- The targets differ from the digits only by being letters. Studies that mark a target by colour, as Raymond et al. did, may give a different curve.
- Sperling's cue was a tone; here it is an arrow, which you have to see first, so his values are only a rough comparison. He also kept the delay fixed for a series of trials, while here the delays are mixed.
- 15 trials per lag give wide intervals, and tiredness near the end of a run lowers the scores; use the break.
- This is not a medical test. This is not a test for attention problems or ADHD.
Controls
- Keyboard: type the letters you saw; Backspace undoes the last one; Enter ends a whole report; Esc pauses. Enter or Space presses the focused button.
- Mouse and touch: pick the letters on the pad that opens over the field; ⌫ undoes the last one; Done ends a whole report. After a pause, a break or a result, buttons react only after 0.7 s, so a double press cannot skip a step.
- Task and Length: in Settings, before a run.
- Run anyway (less precise): see Limits.
- Language: the small link at the end of History switches between English and Slovak.
FAQ
Why did I see only one letter?
Most likely the second one came during your blink, which is the effect being measured. Pick a guess anyway: with 17 possible letters a lucky guess is rare, and an empty answer is not allowed.
Sources
- Raymond J. E., Shapiro K. L. & Arnell K. M. (1992). Temporary suppression of visual processing in an RSVP task: an attentional blink? Journal of Experimental Psychology: Human Perception and Performance 18(3):849–860. doi:10.1037/0096-1523.18.3.849
- Dux P. E. & Marois R. (2009). How humans search for targets through time: a review of data and theory from the attentional blink. Attention, Perception, & Psychophysics 71(8):1683–1700. PMC2915904
- Sperling G. (1960). The information available in brief visual presentations. Psychological Monographs: General and Applied 74(11):1–29. doi:10.1037/h0093759