The anti-saccade task has been used to measure attentional control related to general anxiety but less so with social anxiety specifically. Previous research has not been conclusive in suggesting that social anxiety may lead to difficulties in inhibiting faces. It is possible that static face paradigms do not convey a sufficient social threat to elicit an inhibitory response in socially anxious individuals. The aim of the current study was twofold. We investigated the effect of social anxiety on performance in an anti-saccade task with neutral or emotional faces preceded either by a social stressor (Experiment 1), or valenced sentence primes designed to increase the social salience of the task (Experiment 2). Our results indicated that latencies were significantly longer for happy than angry faces. Additionally, and surprisingly, high anxious participants made more erroneous anti-saccades to neutral than angry and happy faces, whilst the low anxious groups exhibited a trend in the opposite direction. Results are consistent with a general approach-avoidance response for positive and threatening social information. However increased socio-cognitive load may alter attentional control with high anxious individuals avoiding emotional faces, but finding it more difficult to inhibit ambiguous faces. The effects of social sentence primes on attention appear to be subtle but suggest that the anti-saccade task will only elicit socially relevant responses where the paradigm is more ecologically valid.