There is a growing body of research pointing to the possibility that anxious individuals may have difficulty disengaging their attention from threat-relevant information when this information is task irrelevant (e.g., Amir, N., & Elias, J. (2002). Allocation of attention to threat in social phobia: difficulty in disengaging from task irrelevant cues, Manuscript under review; The Quarterly J. Expo. Psycho. 54A (2001) 665). In the current paper, we report a direct test of this hypothesis in individuals with social phobia. Participants performed a variation of the Posner paradigm (Quart. J. Exp. Psycho. 32 (1980) 3). Social threat, neutral, or positive words cued one of two locations on the computer screen. After the cue disappeared, participants had to detect a probe ("(*)") that appeared in one of the two locations. On some trials the cue was valid (i.e., the probe appeared in the same location as the cue). On other trials the cue was invalid (the probe appeared in a different location than the cue). Yet, on other trials, no cue was presented. All participants were slower in detecting probes following invalid cues than probes following valid cues. Furthermore, individuals with social phobia showed significantly longer response latencies when detecting invalidly cued targets than did controls, but only when the probe followed a social threat word. These results suggest that individuals with social phobia may have difficulty disengaging their attention from socially threatening material.