A number of studies using the dot-probe task now report the existence of an attentional bias to angry faces in participants who rate highly on scales of anxiety; however, no equivalent bias has been observed in non-anxious populations, despite evidence to the contrary from studies using other tasks. One reason for this discrepancy may be that researchers using the dot-probe task have rarely investigated any effects which might emerge earlier than 500 ms following presentation of the threat-related faces. Accordingly, in the current study we presented pairs of face stimuli with emotional and neutral expressions and probed the allocation of attention to these stimuli for presentation times of 100 and 500 ms. Results showed that at 100 ms there was an attentional bias towards the location of the relatively threatening stimulus (the angry face in angry/neutral pairs and the neutral face in neutral/happy pairs) and this pattern reversed by 500 ms. Comparisons of reaction time (RT) scores with an appropriate baseline suggested that the early bias toward threatening faces may actually arise through inhibition of the relatively least threatening member of a face pair rather than through facilitation of, or vigilance towards, the more threatening stimulus. However the mechanisms governing the observed biases are interpreted, these data provide evidence that probing for the location of spatial attention at 500 ms is not necessarily indicative of the initial allocation of attention between competing emotional facial stimuli.