Increased vigilance to threat-related stimuli is thought to be a core cognitive feature of anxiety. We sought to investigate the cognitive impact of experimentally induced anxiety, by means of a 7.5% CO(2) challenge, which acts as an unconditioned anxiogenic stimulus, on attentional bias for positive and negative facial cues of emotional expression in the dot-probe task. In two experiments we found robust physiological and subjective effects of the CO(2) inhalation consistent with the claim that the procedure reliably induces anxiety. Data from the dot-probe task demonstrated an attentional bias to emotional facial expressions compared with neutral faces regardless of valence (happy, angry, and fearful). These attentional effects, however, were entirely inconsistent in terms of their relationship with induced anxiety. We conclude that the previously reported poor reliability of this task is the most parsimonious explanation for our conflicting findings and that future research should develop a more reliable paradigm for measuring attentional bias in this field.