本研究的主要研究目的在運用項目反應理論的部皆計分模式和等反應模式於大學聯考，分析考生實際的答題資料，以了解倒扣的計分方式對抑制考生猜測行為的影響。探討的主題有二；(1)運用項目反應理論為考生所估計的能力參數，與運用古典測驗理論為考生所求得的總得分，是否相符。(2)各種題型現行的計分層級是否能抑制猜測行為。研究從八十學年度大學聯考十多萬名考生的實際資料，隨機抽模三千人作為分析的代表性樣本，以英文科作為分析的科目。結果顯示：(1)運用題目反應理論所估計的能力參數，與運用古典測驗理論所求得的總得分，相關甚高；(2)現行的計分層級合理，確有抑制部皆考生猜測行為之功皮。而英文科資料的單一向度性相當良好。在部分計分模型和等級反應模型上，對於大部分的題目而言，資料－模型符合度也不錯。因此將題目反應理論的部分計分模型和等級反應模型，應用於目前大學聯考的考生答題資料分析上。

　　　　　The purpose of the present study is to apply polytomous IRT Models (Partial Credit Model and Graded Response Model) to the guessing-corrected items. The research questions were: (1)Whether the ability estimates resulted from IRT is compatible to total scores resulted from CTT; (2)Whether the scoring system could function as inhibiting guessing behaviors. The test data were randomly selected from the college entrance examinations (N=3000). The results showed that (1)there's high correlation between IRT and CTT regarding ability estimates (2)the current scoring system is appropriate in inhibiting guessing behavior. The general conclusion is that it's appropriate to apply polytomous mode to guessing-item correlation for college entrance examination.