Attentional Bias: What Your Attention Reaches For First

Attention is a selection device. Far more reaches the eyes and ears than can be processed, so something has to decide what gets through, and that decision is made in a few hundred milliseconds, before any deliberate choice. Attentional bias is the name for the systematic part of that selection: the same kinds of information win repeatedly, and which kinds depends on the person and on what has recently mattered to them.

Two mechanisms are usually distinguished. The first is faster orienting: the eyes and the focus of attention arrive sooner at a threatening word, face or object than at a neutral one. The second is slower disengagement: once attention has landed on the threatening thing, it takes longer to leave. The dot-probe task that MacLeod, Mathews and Tata introduced in 1986 indexes the combined pull by timing how quickly a probe is detected in the position where a threat word had just been, but on its own it does not tell the two mechanisms apart: a faster probe response on the threat side is equally consistent with arriving sooner and with leaving later. Koster, Crombez, Verschuere and De Houwer added neutral baseline trials in 2004 for exactly that reason, and with those trials in the design the effect looked at least partly like slower disengagement rather than faster detection.

Priming matters as much as threat. Something looked for, worried about or seen often is easier to notice afterward, which is why a particular car model seems to appear everywhere the week after you consider buying one. That overlaps with the frequency illusion, where the extra noticing is then read as the thing having become more common. Attentional bias is also studied in addiction: Field and Cox reviewed the evidence in 2008 that drug and alcohol cues capture the attention of heavy users, and that the strength of that capture tracks craving.

The honest summary of the evidence has two parts. Bar-Haim and colleagues' meta-analysis of 172 studies in 2007 found a threat-related bias averaging 0.45 standard deviations among anxious participants and nothing significant among nonanxious ones, so this is a reliable group difference rather than something everyone shows. The measurement side is weaker than the phenomenon: Xu and colleagues ran 36 variations of the dot probe with 9,600 participants and reported in 2025 that none of them had internal reliability above zero. The group effect survives; one person's score from one sitting does not mean much.

Colin MacLeod, Andrew Mathews and Philip Tata published "Attentional bias in emotional disorders" in the Journal of Abnormal Psychology in 1986. It introduced the visual dot-probe task, which became the standard way to measure where attention goes under threat.

Participants watched pairs of words on a screen, one above the other, separated by less than two degrees of visual angle, so both fell near the center of vision. One word of each pair was threatening and the other was a neutral word matched to it for length. The threat words covered physical threat, such as injury and agony, and social threat, such as criticized and ashamed. On some trials a small dot appeared in the position where one of the words had been, and the task was to press a key as soon as the dot was detected. The logic is simple: if attention was already at that position, the dot is detected sooner.

Three groups took part: patients with generalized anxiety disorder, patients with depression, and people with neither. The anxious patients detected probes faster when the dot replaced a threat word than when it replaced the neutral word beside it, which indicates that attention had shifted toward the threat. The control participants showed the opposite tendency, slightly faster detection away from the threat word. The depressed patients did not show the anxious pattern, which was the paper's second point: this was not a general feature of emotional distress.

This page does not quote millisecond figures from the 1986 paper because the exact latencies could not be checked against the original text. The direction of the result and the three-group comparison are what the paper is cited for here, and the numeric claim on the demo comes from the 2007 meta-analysis instead.

Attention is pulled before any decision is taken, so the effect shows up as what you noticed rather than as what you chose. The rows below are illustrations, not findings from the studies cited on this page.

Attention is faster than deliberation, so the countermeasures are mostly about the environment and about what happens after the pull, not about trying to not notice.

• Change the field of view rather than the willpower. Moving the phone, the bottle or the notification out of sight removes the cue that wins the competition in the first place.
• Count what attention skipped. If one face, one comment or one symptom has your attention, the useful question is how many comparable things were in the same scene and went unnoticed.
• Put the rate next to the case. A vivid instance holds attention; a base rate has to be written down to compete, which is why the two belong in the same note.
• Expect the second mechanism. Difficulty leaving a threatening thought is the disengagement half of the effect, and setting a later time to return to it works better than arguing with it now.
• Be careful with attention training apps. Attention bias modification, which trains attention away from threat, has produced mixed results in trials, and the reliability problems Xu and colleagues documented in 2025 apply to the tasks these programs are built on.
• Treat a single score as nothing. One sitting with a dot probe, or one set of ratings like the demo above, is too noisy to say anything about a person.

This page is educational. Persistent anxiety, low mood or craving is not something a demo can assess, and a licensed professional is the right person to talk to about it.

Several biases describe what gets noticed, and the differences are in what each one adds after the noticing.

• It is not selective perception, which is about how expectations shape the interpretation of what has already been attended to. Attentional bias is a step earlier: it decides what gets looked at.
• It is not negativity bias, the greater weight given to negative information in judgment and memory. Threat capturing attention is one contributor to it, but negativity bias is about the weighing rather than the looking.
• It is not the frequency illusion, in which a newly noticed thing seems suddenly common. That one takes attentional bias and adds a wrong inference about how often the thing occurs.
• It is not a diagnosis. The group difference in Bar-Haim's meta-analysis was between people with and without anxiety disorders, measured over many trials; it is not a test that sorts individuals, and no demo on this site attempts that.