The training of tactical decision making increasingly occurs through serious computer games. A challenging aspect of designing such games is the modelling of human emotions. Two studies were performed to investigate the relation between fidelity and human emotion recognition in virtual human characters. Study 1 compared five versions of a virtual character that expressed emotions through different combinations of posture, facial expression, and tone of voice. Results showed that emotion recognition was best when all three behavioural cues were present; posture + face and posture + tone of voice were joint second best. In study 2, these three versions were supplemented with contextual information. Cross-variant comparisons yielded marginal differences in emotion recognition and no differences in tactical decision making. Together, these findings suggest that the combination of posture with either facial expression or tone of voice is sufficient to ensure recognition of human emotions in tactical decision-making games.