In this paper, we present a knowledge learning diagnosis approach to supporting a computer-supported collaborative learning environment. Bayesian network technique is used here to diagnose the misconception about learning knowledge and to reason potential misconception for individual learners. After learners have made a test, the system using probabilistic reasoning will automatically create learning communities based on the learners? characteristics and test results. This work is motivated by diagnosing learning performance of learners for both instructors and learners to understand concept comprehension of learners in a computer-supported collaborative learning (CSCL) environment.