This thesis is an investigation of "Using POMDP-based Reinforcement Learning for Online Optimization of Teaching Strategies in an Intelligent Tutoring System". A challenge in building an intelligent tutoring system (ITS) is to create and maintain an optimal teaching strategy. We cast an ITS as a partially observable Markov decision process (POMDP), and apply a reinforcement learning (RL) algorithm to learn the optimal teaching strategy through interactions between the system and the students. The optimal teaching strategy is chosen correctly and efficiently in tutoring a student, it is also learned and maintained in an online model. We present an RL algorithm based on POMDP for learning optimal teaching strategy, then describe the experiments and analyse the experimental results. The experiment has showed that the technique can remarkably improve an ITS's teaching performance.