Anxiety states are associated with increasedattention to threat and a greater likelihood of reachinga pessimistic interpretation of ambiguous events.Existing models of this selective processing possess features that are difficult to reconcile withcurrent experimental findings. In this paper we build onthese earlier ideas to develop a new model,incorporating adaptations that allow it to accountbetter for the accumulating data. Essential featuresare that attributes or meanings of stimuli are processedin parallel and compete for attentional resources. Inputfrom a threat evaluation system (TES) strengthens activation of threat-related attributes, to anextent influenced by anxiety level. Such activation canbe countered, within limits, by voluntary task-relatedeffort, and the balance between these opposing influences determines the extent of anyattentional or interpretative bias seen. Such a model isplausible from an evolutionary perspective and isconsistent with neurological evidence concerning theacquisition and extinction of aversiveconditioning.