Background: Cognitive models of anxiety propose that people with anxiety disorders show elevated levels of attention bias toward threat, but the most commonly used index of attention bias, which measures the construct with an aggregate score of multiple trials across an experimental session, shows poor testretest reliability. Newer indices that measure attention bias dynamically on a trial-to-trial basis show good reliability and enable researchers to measure not only overall attention bias toward threat, but also attention bias variability.Methods: The current study tested the hypothesis that people diagnosed with social anxiety disorder would show higher attention bias variability and higher attention bias toward threat when calculated dynamically and when calculated using the traditional aggregate index. Participants diagnosed with social anxiety disorder (n = 47) and controls (n = 57) completed a 160-trial version of the dot-probe task using emotional and neutral images of faces as stimuli.Results: Relative to controls, participants diagnosed with social anxiety disorder showed higher mean bias toward threat, but only when calculated using trial-level bias scores. There were no differences between groups on attention bias variability.