Collaborative Virtual Environments (CVEs) transform today's computer networks into navigable and inhabited spaces for collaborative work and social play. Various domains have experienced the beneficial use of CVEs, particularly in training and entertainment. Inhabitants are represented as virtual actors, which convey their presence, identity and status of activity, issues that are essential for enhancing social interaction. However, understanding social interaction in CVEs and the way virtual actors could enhance this, is still a challenge for human factors. Various issues complicate such a study: the vast amount of factors involved in the construction of CVEs and virtual actors; the current immaturity of the technology and the prototypical nature of current applications, which restrict the full potential of the technology to be investigated; the lack of a rigorous research methodology for studying and informing CVE design. The research studies the role the virtual actors adopt in CVEs for learning; it provides guidelines for their use in such environments to support social interaction and pedagogical concerns; and directs the underlying CVE technology development to satisfy real user and application needs. To determine requirements for the use of virtual actors in CVEs for learning the research uses as its case study the work of the Manchester Museum Education Service with children at Key Stage Level 2 (-9-11 years old) of the National Curriculum. The learning situation is based on `sent', an ancient Egyptian board game from the Museum's collection of artefacts from the pyramid builders' town of Kahun. The study is structured in three distinct phases of increasing sophistication in terms of: the technology used for building prototypes; the population involved in each phase; and the study of user actions and activities in real life to identify interaction requirements between virtual actors within a CVE. The prototypes are studied with `real users'. This process gradually builds on knowledge of interaction and communication issues that arise in the learning situation. This knowledge forms a set of design guidelines for the use of virtual actors in CVEs for learning. These are implemented using the Deva CVEs system developed by the Advanced Interfaces Group at Manchester University. The prototype is studied with real users. This process refines and extends the design guidelines; and derives technology requirements for the development of the underlying CVE and virtual actors technology by clarifying where the current CVE technology fails to support the design guidelines developed in this research