Among the fe;ming effects which are expected by collaborative learning in this study we concentrate on problem-solving knowledge fixing via 'Learning hy Teaching' and problem-solving knowledge aquisition via 'Learning by Observing'. We propose a virtual collaborative learning environment called CASLE (Companion AgentS Learning Environment), embedding Companion Agents which have each of these learning goals in their Teaching Strategies. These Companion Agents' behavior is based on a 'Student Model', which represents the understanding state of a Real Student, tmd a 'Learning Pattern Model', which represents what level of knowledge carding to each Teaching Strategy, a real student is expected to acquire by experiencing a situation in the collaborative learning environment A Facilitator Agent manages these Models and controls the Companion Agents' behavior. Moreover, it identifies faults of the terming pattern of the real student by comparison with the Learning Pattern Model, and it points then out to him.