In networked virtual environments; when the participants are represented by virtual human figures, the articulated structure of the human body introduces a new complexity in the usage of the network resources. This might create a significant overhead in communication, especially as the number of participants in the simulation increases. In addition, the animation should be realistic, as it is easy to recognize anomalies in the virtual human animation. This requires real-time algorithms to decrease the network overhead, while considering characteristics of body motion. The dead-reckoning technique is a way to decrease the number of messages communicated among the participants, and has been used for simple non-articulated objects in popular systems. In this paper, we introduce a dead-reckoning technique for articulated virtual human figures based on Kalman filtering, and discuss main issues and present experimental results.