Performing under the threat of injury or death is a significant source of stress for military operators; despite these risks maintaining a high level of performance is imperative. Understanding predictive mechanisms underlying performance during high stress is critical for developing adaptive training paradigms and technologies to enhance performance. In our study, trained operators (n = 15) performed a shooting task in a simulator in which they had to discriminate enemy vs. friendly targets and decide to shoot or refrain, respectively. Decision accuracy was measured as the performance outcome. Three levels of stress were induced by manipulating performance feedback on incorrect trials: low (LO; none), medium (MD; visually displayed), high (HI; electric shock). Shock was delivered using a ThreatFire™ belt with a 200ms, 50µA pulse for incorrect decisions. The shock was administered no sooner than every 30s. The voluntary, fully informed consent of the participants was obtained in adherence with the Declaration of Helsinki. EKG and respiratory signals were used to extract features to relate to performance. Four features used were derived in overlapping 35s windows: low frequency heart rate variability (LF-HRV), high frequency heart rate variability (HF-HRV), respiratory rate (RR), and heart rate (HR). Each of these features (and their interactions) were related to performance at various time delays using a generalized linear mixed effects model. The final step was a backward selection procedure to generate the most parsimonious model. Results, reported as Odds Ratio (OR), revealed that in the HI stress condition, performance was worse than in LO and MD stress conditions (OR = 0.82, 95% C.I. 0.72-0.93, p = 0.002). HR alone was inversely related to performance at 2 distinct time periods: 1) coincident with behavior (OR = 0.92, 95% C.I. 0.88-0.98, p = 0.003) and 2) 50-90 s prior to behavior (OR = 0.88, 95% C.I. 0.80-0.96, p = 0.003). The interactions between HR 50-90s prior to behavior in MD and HI stress conditions were positively related to accuracy (MD, OR = 1.16, 95% C.I. 1.01-1.33, p = 0.03), (HI, OR = 1.15, 95% C.I. 1.01-1.32, p = 0.03). Further analysis demonstrated that HR was positively related to LF-HRV in all conditions, however, HF-HRV was positively related to HR only in the MD and HI stress conditions F(1,592) = 5.8, p = 0.02; F(1,599) = 26.5; p = 0, respectively (Figure 1). Interpreting LF-HRV as sympathetic activity and HF-HRV as parasympathetic activity, we suggest an important parasympathetic component may influence performance under stress. Relative to sympathetic activity, parasympathetic activity has little influence on HR but seems to generate a functional difference between the relationship between autonomic function and behavior at various levels of stress. Additional work is needed to understand the biochemical pathway associated with these findings to further explore the complex temporal and biological dynamics.