Journal of Abnormal Psychology 1990, Vol. 99, No. 2, 166-173 Copyright 1990 by the American Psychological Association, Inc. 0021-843X/90/S00.75 Attentional Bias in Anxiety: Selective Search or Defective Filtering? Andrew Mathews, Jon May, Karin Mogg, and Michael Eysenck University of London, England Two experimental tasks were used to investigate the nature of a previously documented bias in attention associated with anxiety. Results from the first task failed to reveal any differences between anxious and nonanxious subjects, either in attention focusing or selective search for letters. The second task, with words as targets and distractors, suggested that selective search was less efficient in anxious subjects when distractors were present. Currently anxious subjects were slower than controls when required to search for the target among distractors of any type, whereas both currently anxious and recovered subjects were slower when the distractors were threatening words. It was therefore suggested that a bias favoring threat cues during perceptual search is an enduring feature of individu- als vulnerable to anxiety, rather than a transient consequence of current mood state alone. It has already been documented that clinically anxious clients and high trait-anxious normal subjects under stress are charac- terized by an attentional bias favoring emotional stimuli (Mac- Leod & Mathews, 1988; MacLeod, Mathews, & Tata, 1986). When subjects are required to name the color in which words are written, the presence of threatening, as opposed to non- threatening, words has a greater slowing effect on anxious sub- jects than on controls (Mogg, Mathews, & Weinman, 1989; Richards & Millwood, 1989). Even when presented outside the focus of attention during dichotic listening, irrelevant threat stimuli can attract processing resources and thereby disrupt task performance (Mathews & MacLeod, 1986). As a result, we have proposed that this bias may have important consequences, in that it seems calculated to maintain and intensify anxietyby directing processing resources toward whatever potential threat cues may exist in the environment (Mathews, 1988). A number of important questions remain unanswered about the mechanisms underlying such a bias and about the nature of the stimuli that provoke it. In most of the above studies, neutral or mildly positive words have been compared only with highly salient and emotionally negative words, such as humiliation or mutilated. It remains to be shown whether or not other types of stimuli, emotional or otherwise, are also capable of attracting disproportionate processing resources in emotionally disturbed people. It is well established that the cognitive efficiency of anxious individuals is more impaired than that of nonanxious subjects as task complexity increases, perhaps because of reduced work- ing memory capacity (for reviews, see Eysenck, 1982, 1984). Consequently, the apparently specific attentional bias for threat could arise in part from a general inability to maintain atten- tional focus and avoid distraction. In the experiment reported The work reported here was supported by grants from the Medical Research Council U.K., and the Welcome Trust. Thanks are due to Colin MacLeod for suggesting the use of the exper- imental paradigm used here. Correspondence concerning this article should be addressed to An- drew Mathews, who is now at the Department of Psychology, Louisiana State University, Baton Rouge, Louisiana 70803. by MacLeod et al. (1986), clinically anxious subjects were faster to detect probes that followed threatening words, which sug- gests that they could not avoid attending to them. However, if anxious subjects have less ability to resist distraction in general, then any salient stimulus may capture their attention despite being irrelevant to the task. Such a possibility could not be as- sessed in that study, as no measure of general distractibility was available. Evidence from some other paradigms argues against such a general degree of distractibility in anxiety. For example, Ehlers, Margraf, Davies, and Roth (1988) confirmed that panic patients were slowed more than controls when color-naming physical threat words, but those researchers found no differential effects in the conventional color-conflict form of the Stroop test (in which color-names are printed in contradictory colors). How- ever, the source of the Stroop interference effect is still poorly understood, and processes other than the control of attention may well be involved. It thus remains possible that anxious in- dividuals have general difficulties in controlling the focus of their attention. If this were the case, it would have important implications for theories that depend on the notion of an atten- tional bias that specifically favors threatening information. In describing the nature of attention, one can draw a distinc- tion between "filtering" (or focused attention), in which early selection is based on some simple physical attribute such as lo- cation, and "selective set" (or search), in which selection is based on the category to which a stimulus belongs (Broadbent, Broadbent, & Jones, 1986). Thus, if the location of a target is signaled in advance, then responding can be based on simple filtering, by maintaining attention on the signaled location. Al- ternatively, if the location is unknown but a particular category is to be detected, then responding must be based on the slower search process. Broadbent et al. (1986) have suggested that the difference between detection times found in tasks that do pro- vide advance knowledge of target location and times found in tasks that do not provides an index of effective filtering. This index is the speeding of detection that results from having atten- tion already focused on the target location, rather than having to carry out a selective search process. In experimental studies of normal subjects, Broadbent et al. (1986) did not find any simple relationship between attention 166