The existence of cognitive biases in anxiety is now well established, and we summarize evidence demonstrating attentional vigilance to cues associated with threat, pessimistic interpretation of ambiguous items and an increased perception of the likelihood of occurrence of negative events. We explore how these reactions can be understood within an evolutionary context, and present a descriptive model consistent with the experimental findings, conducive to modification of responses through learning. A computational implementation of aspects of the model successfully simulates changes in reaction time for a simple task as anxiety levels increase. Future directions include pursuing the causal nature of biases in anxiety and examining the potential for change through training techniques.