In the Collaborative Virtual Environments (CVE), there is a conflict between maintaining the state consistency and supporting more users sharing the virtual environment. A self-adaptive message filtering algorithm based on consistency QoS is proposed to improve both the consistency and scalability. Each user in the virtual environment has active workspace, influence area and dormancy zone according to the different requirement of consistency QoS. Time-space bounding sphere is defined to predict user’s active workspace to provide high consistency QoS. When two users’ time-space bounding spheres overlap, the key updates of two users are sent to each other in time to keep the state consistency. A self-adaptive threshold function model is used when two users’ influence areas overlap, which makes a tradeoff between consistency and scalability and provides a medium consistency QoS. Heart-beat packets are sent when two users’ dormancy zones overlap to filter messages in order to reduce the system traffic since the low consistency QoS is required. The results of traffic test and consistency QoS test in the experiment show that our algorithm can maintain the state consistency and reduce the consumption of network resources simultaneously.