Background/Aims: Anxiety disorders are highly debilitating, pervasive, and moderately heritable, with genetic factors contributing moderately to variance in severity. To probe the threat processing mechanisms contributing to this variance, computational reinforcement learning (RL) models were fitted to differential fear conditioning data, collected remotely via a smartphone app, to estimate trial-by-trial threat learning rates, specifically threat acquisition, safety learning, and threat extinction rates. Across two large, independent samples (n=145 and a twin cohort of n=925), we consistently replicated the finding that a slower rate of threat extinction learning was significantly associated with higher anxiety severity (GAD-7). A twin analysis revealed that the threat extinction rate was mildly heritable (h2=0.21), yet a bivariate analysis demonstrated largely separate genetic influences for this learning rate and general anxiety severity. To further explore genetic effects, in a large genotyped cohort (n = 1113), we will use multiple regression to test if Polygenic Risk Scores (PRS) for anxiety, depression, and neuroticism can predict threat learning rates, offering a novel approach to identify pathways where polygenic risk for affective disorders manifests as measurable, mechanistic deficits in threat processing.