We study the causal nature of the relationship between stress and risk-taking and analyze whether risk taking changes under stress depending on the risk taking behavior when not stressed. We design an experiment in which we observe risk taking choices in the loss and in the gain domain for the same participants in consecutive weeks with and without exogenously induced acute stress. In the stress condition, participants were stressed by the Trier Social Stress Test for Groups, while a control protocol was implemented in the control condition. Our stress manipulation was successful as assessed by salivary cortisol, heart rate, and self-reported stress. Our analysis reveals that the effect of stress on risk-taking depends on the underlying risk-taking tendencies. Individuals make decisions in a more risk-seeking manner on lotteries for which their choices were risk-averse in the control condition; and in a more risk-averse manner on lotteries for which their choices were risk-seeking in the control condition.