Abstract In human conversation, meaning is transported through several channels such as verbal and nonverbal behavior. Certain of these behavioral aspects are culturally dependent. Mutual understanding or acceptance is thus, amongst others, depended on the cultural background of the interlocutors. When designing virtual character behavior, culture should be considered as it may improve the character’s acceptance by users of certain cultural backgrounds. This paper proposes a hybrid approach for the generation of culture-specific behaviors in a multiagent system. A computational model has been established by refining theoretical knowledge of culture-specific behavior with statistical data extracted from a video corpus of German and Japanese first-time meetings. Evaluation studies of such culturally enhanced virtual characters were conducted in both targeted cultures. Results indicate that human observers tend to prefer character behavior that was designed to resemble their own cultural background.