Generalised anxiety disorder (GAD) is a common and distressing mental health problem that is characterised by uncontrollable worry, inattention, and physiological arousal. A better understanding of anxiety disorders could be enabled with effective human models and may improve upon the poor predictive validity of animal models in the identification of treatments. The inhalation of 7.5% carbon dioxide (CO2) produces robust increases in subjective and physiological symptoms of anxiety (Bailey, Argyropoulos, Kendrick, & Nutt, 2005) and impairs attentional control (Garner, Attwood, Baldwin, James, & Munafo, 2011a) in healthy volunteers. Research in animals (Ziemann et al., 2009) has provided evidence for the critical involvement of the amygdala in the detection and coordination of anxious behaviours to CO2, however the role of the amygdala in the human response to CO2 has not been identified. Study One used eye-blink startle magnitude, a well-validated, yet indirect measure of amygdala activity, to examine whether CO2 inhalation in humans might provoke a fear response coordinated by the amygdala. Despite producing strong anxiogenic effects on subjective mood and autonomic arousal, 7.5% CO2 had no effect on startle magnitudes. Instead, CO2 inhalation slowed eye-blink startle latencies. This may be due to fewer available processing resources to respond to an external threat. Study Two investigated the extent to which duloxetine, an effective treatment for GAD, might reduce subjective anxiety and associated deficits in attentional control experienced during 7.5% CO2inhalation. Using the antisaccade task, CO2 inhalation was found to increase the number of antisaccade errors (suggesting poor control of attention) in the placebo group; an effect that was not seen in the duloxetine group. No clear effect of duloxetine on CO2-induced subjective anxiety or physiological arousal was revealed. Study Three extended these findings with the novel drug memantine, a licenced treatment for cognitive degeneration in Alzheimer’s disease that has displayed some preclinical anxiolytic properties (Minkeviciene, Banerjee, & Tanila, 2008). Two week administration of memantine in healthy volunteers resulted in significantly fewer antisaccade errors than placebo, in the absence of any change in subjective mood. Collectively, these results suggest that modelled deficits in attentional control in healthy volunteers can be targeted by known and potential treatments of anxiety disorders, without a comparable decrease in subjective and autonomic symptoms. Examination of symptoms across the anxiety phenotype with human models could help identify better treatments, and understand the neural basis underlying pathological anxiety.