We analyze the impact of stress on risky choice in a large-scale between-subjects design with N = 196 participants. In particular, the main contribution of our work is that we are able to disentangle errors in decision making from an actual shift in preferences between our experimental treatments STRESS and NO-STRESS. Acute stress is induced using the Trier Social Stress Task for Groups (TSST-G, von Dawans et al. (2011)), risk preferences are elicited using a recent method developed by Andersson et al. (2016). Additionally, controlling for cognitive reflection (CRT; Frederick (2005), Toplack et al (2014)), we analyze whether our results are driven by different levels of cognitive ability as the driver for (increased) observed noise between treatments. Our main results show that there is no shift in risky choice towards more or less risky choices in the stress condition. Thus, we do not find a significant effect of a change in risk preferences between treatments. However, in line with previous literature, we find that, on average, a higher score in the cognitive reflection measure leads to significantly less noise in the decisions in the choice task.