Anxiety states are associated with increased attention to threat and a grea
ter likelihood of reaching a pessimistic interpretation of ambiguous events
. Existing models of this selective processing possess features that are di
fficult to reconcile with current experimental findings. In this paper we b
uild on these earlier ideas to develop a new model, incorporating adaptatio
ns that allow it to account better for the accumulating data. Essential fea
tures are that attributes or meanings of stimuli are processed in parallel
and compete for attentional resources. Input from a threat evaluation syste
m (TES) strengthens activation of threat-related attributes, to an extent i
nfluenced by anxiety level. Such activation can be countered, within limits
, by voluntary task-related effort, and the balance between these opposing
influences determines the extent of any attentional or interpretative bias
seen. Such a model is plausible fi om an evolutionary perspective and is co
nsistent with neurological evidence concerning the acquisition and extincti
on of aversive conditioning.