「測驗與評量」一直是教學過程中不可或缺的一環，它們可以反應出學生在學習過程中了解知識的程度，以供教學者掌握學生的學習狀況。另從資訊科技的發展來看，多媒體與網際網路使測驗打破時空的限制並具資源分享的功能。同時，題庫系統有助於使測驗的品質更為理想，但如何結合試題反應理論與題庫系統的建置，使其適用範圍更為廣泛，是本研究的主要研究動機。基於上述研究動機，本研究旨在結合試題反應理論、題庫等化，探討線上題庫等化連結策略之優劣，以提供建置電腦化線上適性測驗系統之基礎條件。具體而言，本研究的主要目的為連結不同時間點所收集的線上題庫，進行題庫等化連結，並比較其連結效益。 以下為本研究主要所獲得的結論： 1.題庫等化轉換常數方法以Mean/Mean和Haebara等方法較佳 在本研究進行等化連結時，共進行等化常數轉換方法中，線上轉換常數方法中平均數/平均數（Mean/Mean）及平均數和標準差法（Mean/Sigma）；以及非線性轉換常數方法中特徵曲線法（Stocking-Lord及Haebara）等四種方法。研究結果顯示：就鑑別度而言，在偏差、均差及均方差方面，以Mean/Mean的方法較佳，其餘三種方法的連結效果相近；就難度而言，在偏差、均差及均方差方面，以Haebara的方法較佳，有7次是差異最小。整體而言，以Mean/Mean和Haebara的連結效果較佳。綜言之，在四種等化連結的方法中，以Mean/Mean和Haebara兩種方法做為題庫等化連結的方法，其效果較為理想，可提供後續探討題庫等化的研究參考。 2.線上測驗與紙筆測驗的試題訊息量相近，但難度偏高 本研究線上測驗的甲乙丙三式題目與九十學年度國中基本學力測驗的結果相較，發現：乙丙兩份試題的訊息曲線形狀與國中基本學力測驗相似，表示兩類測驗所提供的訊息量相近，並無太大的差別。有明顯分別的是，甲乙丙三式的題目訊息曲線都較基本學力測驗右偏，顯示學生在回答線上測驗有較困難的情形，此種現象應是線上題目施測時，答題者無法在電腦上直接作答，還需要將幾何圖形畫到紙本上，在紙本上演算後，再回答線上試題，故就步驟而言，造成學生加倍的負擔，導致訊息曲線呈現右偏的情況。 3.線上測驗連結效益良好 本研究在連結方面發現，線上測驗所收集到的反應資料有四分之三以上皆具有連結的效果，亦即「線上題庫與適性測驗整合系統」所收集到的三式反應資料的試題，對於國中基本學力的測驗題庫具有實質上的效果。

　　　　　“Tests and Evaluation” have occupied a vital corner in education field. In terms of the tests and evaluation, students' learning and knowledge understanding can be reflected and the results can be provided to educators who, therefore, can control students' learning condition. Besides, in the perspective of the development in information technology, thanks to the multimedia and internet which have broken down the confinement of time and space and offered a variety of shared resource, the item bank systems, thus, can facilitate test quality. Therefore, how to connect item response theory with the construction of item bank systems and then how to widen up the application range should deserve further study. Consequently, in terms of the combination of item response theory and item bank equation, the purpose of the study is to discuss the strategic advantages of on-line item bank equation, and to provide basic constructive condition for the computerized system of on-line adaptive testing. Concretely speaking, the study aims at linking item banks collected from different periods of time, equating these linked item banks, and then, comparing the linking performances. The conclusions of this study can be summarized as follow: 1. Concerning the constant transformation of item bank equation, Mean/Mean Methods and Haebara Methods are better options. During the equating and linking, four methods are applied in this study: from linear constant transformation methods, Mean/Mean and Mean/Sigma Methods are presented; from the non-linear methods, two kinds of Characteristic Curve Methods are presented and they are Methods of Stocking-Lord and Haebara. The study shows: concerning discrimination, Mean/Mean Methods display better result in deviation, average deviation and mean square deviation while linking results of the other three methods are similar to each other. Concerning difficulties, Haebara Methods achieve better performance--seven of the results perform the minimal deviation. Generally speaking, Mean/Mean Methods and Haebara Methods have better linking performance. To sum up, among the four equating and linking methods, Mean/Mean and Haebara Methods perform better in the linking of item bank equation. This can be a general reference to those who are interested in item bank equation research. 2. Item information between on-line tests and paper-pencil tests is similar but indicates more difficulties. Three types of on-line test items, Item A, B, and C, and one test item from the Basic Competency Test (BCT) for junior high students in 1991 are employed in this study. The findings are:Information curves of Item B and C are similar to that of the BCT item for junior high students. These items all provide similar test information and therefore, no significant differences can be found. However, information curves of Item A, B, and C are right skewed while that of BCT item for junior high students is not. The result suggests that students have difficulties in on-line tests. The possible explanation is while doing the on-line items, only after the students have drawn the geometric diagram and finished the calculation on their paper can they proceed questions on computers. That is, students are not able to start their on-line tests only by computer. In consequence, the complicate solution process doubles students' burden. This is why the information curve of on-line item is right skewed. 3. Linking performance of on-line test is good The study displays that three-fourths of the responded data collected from on-line tests show linking performance. That is, three responded tests collected from the integrated internet system of on-line item bank and computerized adaptive testing prove to have practical impact on the BCT item banks for junior high school students.