The inclusion of additional modalities into the communicative behavior of virtual agents besides speech has moved into focus of human-computer interface researchers, as humans are more likely to consider computer-generated figures lifelike when appropriate nonverbal behaviors are displayed in addition to speech. In this paper, we propose a knowledge-based approach for the automatic generation of gesture animations for an articulated figure. It combines a formalism for the representation of spatiotemporal gesture features, methods for planning individual gestural animations w.r.t. to form and timing, and formation of arm trajectories. Finally, enhanced methods for rendering animations from motor programs are incorporated in the execution of planned gestures. The approach is targetted to achieve a great variety of gestures as well as a higher degree of lifelikeness in synthetic agents.