The major purpose of this research is to establish a computerized adaptive diagnosis test items and remedial instruction module by investigating elementary mathematics “graphic area,” which includes rectangle, square, parallelogram, triangle, and trapezoid as an example. The research applies Bayesian networks as modeling assessment data combining knowledge structures of experts to establish the module. When the students were tested, the module can identify the skills students have acquired as well as their common error types. It also offers immediate feedback and adaptive remedial teaching by computerized animation, in the hope to provide functions of assessment, diagnosis, and remedial education simultaneously. The results show: 1. The Bayesian networks evaluation mode has applied effectively to the diagnosis of students’ common errors and sub-skills. 2. By integrating the Bayesian networks, knowledge structure of experts, and knowledge structure of students, students' skill indicators, sub-skills, and common error types can be identified more accurately. 3. The progress of students is significant after taking the Computerize adaptive remedial instruction. 4. The similarity between computerized adaptive diagnosis and Bayesian networks stands for high degree.
