To accurately analyze the problems of students in learning, the composed test sheets must meet multiple assessment criteria, such as the ratio of relevant concepts to be evaluated, the average discrimination degree, difficulty degree and estimated testing time. Furthermore, to precisely evaluate the improvement of student's learning performance during a period of time, a series of relevant test sheets need to be composed. In this paper, a particle swarm optimization-based approach is proposed to improve the efficiency of composing near optimal serial test sheets from very large item banks to meet multiple assessment criteria. From the experimental results, we conclude that our novel approach is desirable in composing near optimal serial test sheets from large item banks and hence can support the need of evaluating student learning status. (Contains 4 tables and 6 figures.)