We compare two alternative measures for assessing people’s emotional reactions to political stimuli: the traditional self-report measure and facial expression analysis. We recruited participants to take part in a study examining reactions to a set of negative political commercials aired during the 2018 elections. We compare people’s self-reporting of their emotional reactions to negative political advertisements with their expressed emotion, according to the facial expression analysis. We find the discriminant validity of the facial expression analysis is higher than the self-report measure. Second, the self-report and facial expression measures of emotion have little convergent validity: we fail to find a consistent and strong positive correlation between the self-report and facial software measures of the same emotion and the same political advertisement. Third, the facial software measure has better predictive validity than the self-report measure, generating better predictions for the three dependent variables examined: changes in political interest, changes in people’s confidence in elected officials, and people’s assessment of the tone of the senate campaign.