Affect Control Theory (ACT; Heise 1979, 2007) states that people control social interactions by striving to maintain culturally shared feelings about the situation. The theory is based on mathematical models of language-based impression formation. In a laboratory experiment, we tested the predictive power of a new German-language ACT model with respect to actual behavior and felt emotions in leadership; 60 subjects managed a computer simulated company by communicating with 3 different virtual employees (within-subjects manipulation). Half of the subjects were primed with the concept of authoritarian leadership using a situational interview technique; the remainder was primed with the concept of democratic leadership (between-subjects manipulation). There were 14 dependent variables (leadership categories like praise, criticize, augment salary, etc.). The German impression-formation model correctly predicted 27 of 42 between-subjects contrasts (p less than 0.05) and 56 of 84 within-subjects contrasts (p less than 0.01). Moreover, Euclidean distances of emotions predicted by the German ACT model correlated negatively with the frequency with which the subjects experienced these emotions (correlations ranged from r = -0.18 to r = -0.61). These results support Affect Control Theory's proposition that realistic social interaction can be predicted by mathematical models of affective consistency. (Contains 3 figures, 3 footnotes and 4 tables.)