The main purpose of this research is to develop a digital individual teaching material for “Exposition” teaching of Language Domain in elementary schools. We analyze the knowledge structure of teaching material content, and set the examination paper based on Bayesian Networks and the Competence Indicators. We then input into the computerized diagnosing test system, and proceed with tests for finding out students’ mistake types. It offers adaptive remedy instructions by the mistake types after taking the test online. There are two major stages to complete this study. In the first round data collection, there were 273 fifth-grade students ( from 10 elementary schools in four counties of Central Taiwan ) participated in the study. We took 100 students in Taichung City to be our research subjects in the second round. Some findings are briefly outlined as follows: 1. The Cronbach α value of formal questionnaire is 0.7795. Most of the questions have an excellent discrimination. 80％ of the questions are ranged in a medium or low difficulty level. The number of items administrated by BNAT is 11 or 12 items are omitted and prediction accuracy can reach 97％. 2. After taking the instruction, students have significant progress on their average grades. The teaching model of one v.s. one and one v.s. two are better than one v.s. a class. It shows that this digital individual remedial teaching material is useful. 3. This digital individual teaching materials made by knowledge structure is adaptable to three kinds of teaching models. 4. The Effects of Remedial Instruction for the three kinds of teaching models are very good, especially for the low proficiency learners group. The grade-point average of three kinds of teaching models improved by over 10 points.
