In a near miss experience, a person learns anxiety despite the fact that the feared outcome does not actually occur. Theories based on S-R links or direct encoding of emotional memories do not provide a good model of near miss learning because they fail to account for informational variables. An alternative cognitive model is outlined, based on expectancy of an inferred negative outcome. The model is applied to posttraumatic stress disorder, in which the critical feared outcome is often death, and panic disorder, in which a near miss inference may protect expectancies of harm from disconfirmation. Unlike direct emotional encoding models, the inferential/expectancy model provides multiple cognitive targets for treatment. Near miss learning may be usefully distinguished from other pathways to anxiety, and warrants further experimental investigation.