The Attentional Blink

Abstract

The attentional blink is a brief failure of awareness in which the second of two targets in a rapid stream of items goes unreported when it appears roughly 200 to 500 milliseconds after the first. It is measured with rapid serial visual presentation, in which letters or words appear one at a time at fixation and the observer reports two designated targets. Accuracy for the second target falls sharply at short lags and recovers within about half a second, tracing a U-shaped curve, though a target immediately following the first is often spared. The dominant two-stage account attributes the deficit to a capacity-limited stage of consolidation into working memory that handles only one target at a time. Three interactive demonstrations model the blink curve, the two-stage bottleneck, and the factors that deepen or lift the deficit.

Keywords: rapid serial visual presentation, temporal attention, lag-1 sparing, conscious access

The attentional blink is the finding that once attention has been committed to identifying one target, a second target arriving a fraction of a second later can pass through the visual system without ever reaching report (Raymond et al., 1992). The stimuli are presented and even registered by the senses, yet the observer, asked moments later what the second target was, has no answer. Because the two targets are separated in time rather than space, the effect exposes a limit not on where attention can be aimed but on how quickly it can be redeployed, and it has become one of the most productive tools for studying the boundary between what is merely seen and what reaches conscious perception. This article traces the blink from the presentation method that reveals it, through the sparing that constrains any account of it and the two-stage model that dominates the field, to what determines its depth, the theories that compete to explain it, and its neural signature.

What the Attentional Blink Is

The attentional blink is a temporary suppression of the ability to identify a second target when it appears shortly after a first one in a stream of rapidly presented items. When two targets are separated by roughly 200 to 500 milliseconds, the first is reported well but the second is frequently missed, even though the same second target is reported easily when it appears alone or after a longer gap (Raymond et al., 1992). The name captures the analogy with an eye blink: attention closes for a brief interval after it engages the first target, and events falling within that interval are lost, not because the eyes moved or the stimulus was degraded but because a central process was occupied.

The effect is a limitation of attention over time rather than over space. Classic paradigms show that attention can be aimed at only so many locations at once; the blink shows that it can also identify only so many things in quick succession, because committing to one target imposes a refractory period on the next (Shapiro et al., 1997). The phenomenon was anticipated by earlier work on the identification of multiple targets in fast streams, which found that reporting one item impaired the report of a closely following item (Broadbent & Broadbent, 1987), and it was pattern rather than sensory processing that mattered, since a target defined by its meaning or shape produced the deficit whereas a simple sensory detection did not (Shapiro et al., 1994). The blink is therefore best understood as a bottleneck in the higher-level processing that turns a registered stimulus into a reportable one.

The RSVP Paradigm

The blink is produced with rapid serial visual presentation, in which a sequence of items, typically letters or words, is shown one after another at a single location at a rate of about ten per second, so that each item lasts roughly one hundred milliseconds. Two of the items are designated targets: the first target, conventionally called T1, might be the only letter in a stream of digits, and the second, T2, might be a specified probe such as the letter X whose presence must be reported (Raymond et al., 1992). Because the items appear at fixation and in immediate succession, no eye movement can help, and the observer's task is purely one of extracting two targets from a temporal stream.

The key measurement is the accuracy of reporting T2, given that T1 was reported correctly, plotted as a function of the lag between them, where lag is the number of stream positions separating the two targets and each position adds about one hundred milliseconds. This conditional measure, T2 accuracy given correct T1, is what isolates the blink from a general failure to attend (Chun & Potter, 1995). Figure 1 shows the characteristic result: accuracy is high at the shortest lag, drops to a trough around lags 2 and 3, and climbs back to baseline by about lag 7. The demonstration that follows lets the reader select a lag and a difficulty for the first target and read off the modeled second-target accuracy and the size of the blink.

Lag-1 Sparing

The single most constraining fact about the blink is that it does not begin at the shortest lag. A second target presented in the very next stream position after the first, at lag 1 with no intervening item, is usually reported as accurately as one at a long lag, even though it falls squarely inside the interval where the deficit is otherwise deepest (Chun & Potter, 1995). This lag-1 sparing means the function is not a simple decay that is worst immediately after T1 and eases with time; it is a dip that has to recover on both sides of a spared point, which rules out any account in which the first target simply shuts attention off for a fixed window.

The standard interpretation is that both targets can be admitted together when nothing separates them. On this view a brief window of enhanced processing opens when T1 is detected, and a T2 that arrives while the window is still open is processed as part of the same attentional episode and consolidated alongside T1 (Shapiro et al., 1997). An intervening distractor closes the window, so that by lag 2 the second target arrives after it has shut and must wait for a busy stage to free up. Sparing therefore tells us that the bottleneck is not about the passage of time as such but about whether the second target can join the first before consolidation begins, and it is the phenomenon against which every model of the blink is tested.

The Two-Stage Model

The dominant account of the blink is the two-stage model. In its first stage, every item in the stream is processed in parallel and to a fairly high level, so that targets are detected and given a short-lived, fragile representation that will decay unless it is stabilized. In the second stage, that fragile representation is consolidated into durable working memory, from which it can be reported, but this stage is capacity-limited and can accept only one item at a time (Chun & Potter, 1995). When T1 seizes the second stage, T2 is left waiting in the fragile first-stage buffer, and if consolidation of T1 takes longer than T2 can survive, T2 decays and is lost.

This architecture explains the shape of the curve directly. At short lags the second stage is still occupied with T1, so T2 must wait and is likely to decay before its turn, producing the trough; at long lags T1 has been consolidated and released before T2 arrives, so T2 is processed normally. Lag-1 sparing follows if a T2 arriving immediately can be consolidated together with T1 as a single episode before the bottleneck closes. Response selection and the demands of the first task feed into this picture, because making T1 harder to process, or requiring a speeded decision about it, lengthens the second stage's occupation and deepens the blink (Jolicoeur, 1998). The next demonstration implements the model as a decaying buffer: the reader sets how long T1 takes to consolidate and how fast the T2 trace decays, and the resulting accuracy curve reproduces both the trough and the sparing.

What Determines the Blink

The depth of the blink is not fixed but varies systematically with the stimuli and the task, and those dependencies are among the strongest constraints on theory. Similarity is central: the more the targets resemble the distractors that surround them, and the more the two targets resemble each other, the deeper the blink, because a stronger competing representation is harder to select against (Raymond et al., 1995). Masking matters too, since a second target that is followed by an item, and so is masked, blinks, whereas a T2 that ends the stream with nothing after it is largely spared, indicating that the deficit acts on a representation that a trailing mask can erase before it is consolidated.

The demands imposed by the first target are the other major lever. A first target that is easy to identify produces a shallow blink, while one that is degraded, ambiguous, or requires a speeded response holds the bottleneck longer and produces a deep one (Jolicoeur, 1998). These factors combine roughly multiplicatively, so that an easy, distinctive first target with dissimilar distractors may yield almost no blink at all. The demonstration that follows lets the reader toggle first-target difficulty, target-distractor similarity, and a concurrent distracting task, and watch the modeled second-target accuracy at the trough rise and fall as each factor is changed.

Theories of the Blink

Although the two-stage model organizes most of the evidence, several accounts compete over what the bottleneck actually is. Interference theories hold that the deficit arises not from a strict serial stage but from competition among items in working memory: T1, T2, and the distractors between them are all encoded to some degree, and their overlapping representations interfere, with T2 the likeliest casualty because it is weakest (Shapiro et al., 1994). Delayed-engagement accounts locate the problem earlier, arguing that the intervening distractor after T1 causes attention to disengage, so that when T2 arrives, attention has to be re-engaged and the delay, not a full stage, is what costs the report (Nieuwenstein et al., 2005). Computational models such as the simultaneous-type serial-token account render the two-stage idea precise, separating the parallel typing of items from the serial binding of each to a token that makes it reportable, and reproduce sparing and the trough from that distinction (Bowman & Wyble, 2007). Table 1 sets the leading accounts side by side, and a broad review concludes that no single mechanism captures every result, so that the blink is likely a joint product of a capacity limit and the control that governs it (Dux & Marois, 2009).

The Neural Basis and Conscious Access

Electrophysiology locates the blink at a late, post-perceptual stage. When a second target is missed during the blink, the early sensory components of the event-related potential are preserved, and the target still evokes the N400, a marker that its meaning was extracted; what disappears is the P3, the component tied to updating working memory (Vogel et al., 1998). The blinked target is thus processed all the way to the level of meaning and then fails to be consolidated into the durable, reportable state, exactly the dissociation the two-stage model predicts, and strong evidence that the loss is one of access rather than perception.

The access is close to all-or-none. Using targets whose visibility can be titrated, magnetoencephalography shows that during the blink a second target's fate is bimodal: it either evokes a full, late ignition of frontoparietal activity and is seen, or it does not and is missed, with few intermediate states, so consciousness of the target behaves like a threshold event rather than a gradual fading (Sergent et al., 2005). The structures involved place the blink within a general account of the brain's capacity limits, in which a shared frontoparietal network can grant only one stimulus at a time full access to awareness and action, tying the temporal bottleneck of the blink to the same central limit that produces the dual-task delay of the psychological refractory period (Marois & Ivanoff, 2005).

Overinvestment and Loss of Control

A striking line of evidence suggests the blink is not a hard capacity ceiling but a cost of the way attention is deployed. If the deficit were simply a stage being full, taxing the observer further should worsen it; instead, giving people a concurrent, task-irrelevant activity, such as listening to music or being encouraged to let the mind wander, reliably reduces the blink (Olivers & Nieuwenhuis, 2005). The overinvestment hypothesis reads this as evidence that too much focused effort on the stream is counterproductive: an observer who tries hard exerts tight, item-by-item control that overreacts to the distractor following T1 and suppresses the very stream that contains T2.

The loss-of-control account frames the same idea mechanistically. On this view the distractor after T1 triggers a transient reconfiguration of the attentional filter, an endogenous act of control that temporarily excludes input, and the second target is caught in the resulting gap (Di Lollo et al., 2005). What looks like a passive bottleneck is then partly an active, if maladaptive, gating decision, which is why manipulations that loosen control, whether distraction, a diffuse attentional set, or simply not trying to isolate each target, lift the deficit. These findings do not overturn the two-stage model so much as qualify it: the consolidation limit is real, but how much of it is expressed on any trial depends on the control regime the observer adopts (Dux & Marois, 2009).

Applications

As a laboratory measure, the blink provides a clean index of the temporal limits of attention that is sensitive to clinical and individual differences. Because it isolates the interval over which a single act of target selection ties up central resources, its magnitude and time course vary with conditions that affect attention and working memory, and the RSVP task has been used to probe temporal attention in aging, mood, and disorders of attention (Martens & Wyble, 2010). The conditional logic of the measure, scoring the second target only when the first was caught, makes it a relatively pure assay of the second stage rather than of general alertness.

Beyond the clinic, the blink matters wherever a person must monitor a rapid stream for more than one event. Any display in which important signals can arrive within a few hundred milliseconds of one another, from baggage screening to the reading of instruments, risks a missed second signal for reasons that have nothing to do with its visibility (Marois & Ivanoff, 2005). Understanding that the loss is one of consolidation, and that it can be eased by a less tightly focused attentional set, informs how such tasks are paced and how operators are trained, and it cautions against assuming that a signal which was clearly presented was necessarily perceived.

What the Paradigm Does and Does Not Show

The blink is a robust and well-specified effect, but several of its interpretations remain contested. The chief caution is that lag-1 sparing, and the finding that sparing can extend across two or three items when no distractor intervenes, sits awkwardly with the simplest serial-bottleneck story, and has driven the field toward models in which the boundaries of an attentional episode, not a fixed refractory stage, determine what is spared and what is lost (Nieuwenstein et al., 2005). A model that predicts the trough but not the exact conditions of sparing is incomplete, and no single account yet predicts all of the sparing, masking, and similarity effects together (Dux & Marois, 2009).

A second caution concerns what a missed second target means. That the meaning of a blinked target is still extracted, shown by a surviving N400, establishes that the deficit is not a failure to perceive, but it also warns against treating the blink as a simple on-off gate on awareness (Vogel et al., 1998). The effect measures the probability that a second target is consolidated into a reportable state, and that probability is shaped by control, expectation, and effort as much as by an inherent limit (Martens & Wyble, 2010). The RSVP paradigm is thus a precise instrument for detecting that consolidation failed, and a blunter one for deciding whether the cause on a given trial was a full stage, active suppression, or a lapse of engagement.

Worked Example

Consider the empirical curve behind the first demonstration, with its default constants. The single-target baseline accuracy, the level reached when the second target is easy to report, is 92%. The blink deficit at a given lag is modeled as a bell-shaped loss centered on lag 2, so at lag 2 the deficit is at its full default value of 37 percentage points and the accuracy is 92 minus 37, or 55%. The blink magnitude, the number most often reported, is the baseline minus the accuracy at the trough, 92 minus 55, or 37 points. By lag 4 the deficit has shrunk to about 20 points, giving an accuracy of roughly 72%, and by lag 7 it is under one point, so accuracy has returned to baseline. Lag 1 is treated separately as the spared position, held near baseline at 89%, which is why the lowest point of the curve is at lag 2 rather than at the shortest lag.

The second demonstration reaches a similar curve from a mechanism rather than a shape. Each stream position lasts a stimulus onset asynchrony of 90 milliseconds, the first target takes 400 milliseconds to consolidate, and a waiting second target decays with a time constant of 250 milliseconds. At lag 2 the second target arrives 2 times 90, or 180 milliseconds, after the first, so it must wait 400 minus 180, or 220 milliseconds, for consolidation to finish. Its surviving fraction is the exponential of negative 220 divided by 250, which is the exponential of negative 0.88, or about 0.41. With a guessing floor of 25% and a ceiling of 92%, the predicted accuracy is 25 plus 0.41 times the 67-point range, or about 53%, close to the empirical trough. At lag 5 the arrival time of 450 milliseconds already exceeds the 400-millisecond consolidation time, so the wait is zero, survival is one, and accuracy returns to the 92% baseline. The two demonstrations thus agree at the trough and in their recovery, one by describing the curve and the other by generating it.

Discussion

The attentional blink earned its central place because it made a limit on the timing of awareness measurable. The claim that attention needs time to move from one act of identification to the next became testable the moment the RSVP task attached a report probability to the interval between two targets, and the orderly U-shaped curve that resulted, with its stubborn sparing at lag 1, has disciplined theory for three decades. The two-stage model, the interference and delayed-engagement alternatives, and the electrophysiological localization of the deficit to a late, post-perceptual stage are among the more reproducible and generative findings in the study of attention (Martens & Wyble, 2010).

The trajectory of the field shows the familiar movement from a single mechanism toward an interaction of several. The early picture of a stage that fills and empties has given way to one in which a real consolidation limit is expressed through an active control regime that can overreact to distraction, so that the same observer blinks more or less depending on how tightly attention is held. That the deficit spares the meaning of the missed target, and that it can be reduced by distraction and tied to the brain's general capacity limit, reframes the blink not as a gate that closes on perception but as a bottleneck on the consolidation that turns a perceived thing into a reportable one. Its value lies precisely in how finely that bottleneck, and the control that governs it, can be separated and measured.

Glossary

Attentional blink.

The transient failure to report a second target that appears roughly 200 to 500 milliseconds after a first target in a rapid stream, despite the target being clearly presented.

Bottleneck.

A processing stage that can handle only one item at a time, forcing later-arriving information to wait and, in the blink, to decay before it is admitted.

Consolidation.

The second-stage process that stabilizes a fragile perceptual representation into a durable working-memory state from which it can be reported.

Distractor.

A non-target item in the stream; the distractor immediately following the first target is critical, as its removal produces sparing and its presence helps trigger the blink.

Lag-1 sparing.

The accurate report of a second target that immediately follows the first with no intervening item, despite falling inside the blink interval.

Lag.

The number of stream positions separating the two targets, each adding about one hundred milliseconds; the horizontal axis of the blink curve.

Loss-of-control account.

The view that the distractor after the first target triggers a transient reconfiguration of the attentional filter that mistakenly suppresses the second target.

Masking.

The disruption of a target's representation by an item that follows it; an unmasked final-position target is largely spared, showing the deficit acts before consolidation.

Overinvestment hypothesis.

The proposal that excessive focused effort on the stream deepens the blink, so that distraction or a looser attentional set paradoxically improves second-target report.

Rapid serial visual presentation.

A method in which items are shown one at a time at a single location at about ten per second, used to place two targets close together in time.

Stimulus onset asynchrony.

The interval between the onsets of successive stream items, about one hundred milliseconds in a typical blink task, converting lag into elapsed time.

Target.

An item the observer must report; the first is conventionally called T1 and the second T2, and the blink is scored as T2 accuracy given correct T1.

Temporal attention.

The allocation of attention across time to select events from a stream, contrasted with spatial attention, which selects across locations.

Two-stage model.

The account in which items are detected in parallel at a first stage but consolidated one at a time at a capacity-limited second stage, the dominant explanation of the blink.

Working memory.

The durable, capacity-limited store into which a target must be consolidated to become reportable; the resource the second stage competes for.

Key Researchers

Jane E. Raymond. Professor Emeritus of Psychology at the University of Birmingham; lead author of the 1992 study that named the attentional blink and showed that a masked first target opens a several-hundred-millisecond window in which a second target is missed. ORCID - Faculty Page - Google Scholar

Kimron L. Shapiro. Professor Emeritus of Psychology at the University of Birmingham; co-discoverer of the blink and author of the interference account, arguing that the second target is lost where its representation competes with the first for consolidation. ORCID - Faculty Page - Google Scholar

Karen M. Arnell. Professor of Psychology at Brock University; co-author of the founding 1992 report and of the studies establishing that pattern rather than sensory processing, and target-distractor similarity, drive the blink. Faculty Page - Google Scholar

Marvin M. Chun. Richard M. Colgate Professor of Psychology at Yale University; with Potter formulated the two-stage model in which targets are detected in parallel but pass one at a time through a capacity-limited second stage, the dominant framework for the blink. ORCID - Faculty Page - Google Scholar

Mary C. Potter. Professor Emerita of Psychology at the Massachusetts Institute of Technology; pioneer of rapid serial visual presentation and co-author of the two-stage model, whose work on conceptual short-term memory frames what the blink interrupts. Faculty Page - Wikipedia

Rene Marois. Professor of Psychology at Vanderbilt University; co-authored the standard modern review of the blink and, through neuroimaging, tied it to a central bottleneck of information processing shared with other capacity limits of the brain. ORCID - Faculty Page

Frequently Asked Questions

What is the attentional blink?
It is a brief inability to report a second target when it appears about 200 to 500 milliseconds after a first target in a rapid stream of items, even though the second target is clearly presented (Raymond et al., 1992).

Why is it called a blink?
The name draws an analogy with an eye blink: attention seems to close for a few hundred milliseconds after it engages the first target, and events arriving in that interval are lost, though the eyes never actually close (Shapiro et al., 1997).

What is rapid serial visual presentation?
It is a method in which items such as letters or words are shown one at a time at a single location at roughly ten per second, so that two targets can be placed only a fraction of a second apart (Raymond et al., 1992).

What is lag-1 sparing?
It is the finding that a second target immediately following the first, with no item between them, is usually reported accurately even though it falls inside the blink interval, which constrains every theory of the effect (Chun and Potter, 1995).

How long does the attentional blink last?
The deficit is deepest about 200 to 300 milliseconds after the first target and recovers to baseline within roughly half a second, tracing a U-shaped curve across the lag between the targets (Chun and Potter, 1995).

What causes the attentional blink?
The dominant two-stage account holds that consolidating the first target into working memory occupies a capacity-limited stage, leaving the second target to decay while it waits for that stage to free up (Chun and Potter, 1995).

Can the attentional blink be reduced?
Yes; giving observers a concurrent, task-irrelevant activity or encouraging a looser attentional set reduces the blink, suggesting that too much focused effort deepens it (Olivers and Nieuwenhuis, 2005).

How is the attentional blink related to the psychological refractory period?
Both reflect a central capacity limit that lets only one stimulus at a time reach full processing, and neuroimaging ties the blink and the psychological refractory period to a shared frontoparietal bottleneck (Marois and Ivanoff, 2005).

References