This research is aimed to establish On-Line Diagnostic Test, based on the unit of the girth and area of a fan-shaped in the 6th grade math curriculum. We can diagnose the effectiveness of students’ learning and facilitate self-initiated learning through the five Bayesian networks constructed by expert knowledge structure and Adaptively Remedial Instruction based on combining Multiple Bayesian Networks modules. Moreover, students can follow the pattern and master the skills by repeated practice. This study first analyzes the target content and the wrong types, establishes ability index of expert knowledge structure, then propounds and carries out the pen-and-paper diagnostic test. After the test, the procedures of Adaptively Remedial Instruction are founded, based on combining Multiple Bayesian Networks, in order to evaluate the effectiveness of Adaptively Remedial Instruction. This paper identifies the followings: 1. The identification accuracy of combining Multiple Bayesian Networks is much higher than single Bayesian Networks. 2. The average number of proposition of the Adaptively Diagnostic Test Instruction is 14.31, which is 10.69 lower than pen-and-paper test. 3. Through Adaptively Remedial Instruction, students’ average scores improve significantly. In conclusion, the On-Line Diagnostic Test and Adaptively Remedial Instruction of the girth and area of a fan-shaped unit in elementary school curriculum, based on combining Multiple Bayesian Networks, which are proposed in this study, are proved to obtain better classification results.