An attentional bias to threat is an important maintaining and possibly aetiological factor for social anxiety. Despite this, little is known about the underlying mechanisms of threat biases, such as the relative contributions of top-down and bottom-up attention. In order to measure attentional bias toward threat, the current study employed a variation of the dot-probe task in which participants' attention was initially cued to the left or right side of the screen before an angry face paired with a neutral face was displayed, and subsequently participants responded to a probe in the locus of one of the faces. This design provides separate measures of engagement with and disengagement from threat. In addition, in order to manipulate the availability of top-down attentional resources, participants completed this task under no, low (simple arithmetic task), and high (difficult arithmetic task) working memory load. Higher levels of social anxiety were found to be associated with increased engagement with threat under no-load, whereas this effect was eliminated under low-load and high-load conditions. Moreover, social anxiety was not associated with delayed disengagement from threat. These results highlight the critical role of top-down attention for engaging attention with threat.