Political scientists frequently wish to test hypotheses about the effects of specific emotions on political behavior. However, commonly used experimental manipulations tend to have collateral effects on emotions other than the targeted emotion, making it difficult to ascribe outcomes to any single emotion. In this letter, we propose to address this problem using causal mediation analysis. We illustrate this approach using an experiment examining the effect of emotion on dyadic trust, as measured by the trust game. Our findings suggest that negative emotions can decrease trust, but only if those negative emotions make people feel less certain about their current situation. Our results suggest that only anxiety, a low-certainty emotion, has a negative impact on trust, whereas anger and guilt, two emotions that differ in their control appraisals but induce the same high level of certainty, appear to have no effect on trusting behavior. Importantly, we find that failing to use causal mediation analysis would ascribe a positive effect of anxiety on trust, demonstrating the value of this approach.