本文包括兩個研究，第一個研究乃是關於電腦化適性測驗中(CAT)起始點及記分方法的模擬研究，其結果顯示假如有足夠的訊息可以作為考生能力分配界的基礎時，貝氏計分法是優於最大概率估計法，將其所界定的先前分配的標準差從1提昇到2對貝氏估計法中國歸平的問題有相當的助益。當能力所在的點沒有太多訊息時EAP(expected a posterior)的估計優於MAP(maximum a posterior)的估計，當能力所在的合具備充分的訊息時，MAP及EAP的估計準確性非常接近，MAP有優於EAP的趨勢，隨機起始點的估計準確性，略遜於平均數起始點的結果，但其影響程度並不嚴重。第二個研究主要針對第一個題庫的限制，探討當題庫擴大時上述研究結果的推論是否依然有效。其結果顯示當題庫擴大或訊息提昇時各計分方法間的差異較不明顯，但整體趨勢仍與第一個研究結果相似。當理想題庫存在時，記分方法的選擇可能並不重要，但在實際應用時EAP記分法加上先前分配標準差為2的組合仍是最佳選擇。

　　　　　This research is concerned with the effects of scoring method, entry level, and item pool to the ability estimation precision of computerized adaptive testing (CAT). Two studies were implemented. The results of study I (English test, 100 items) suggest that Bayesian scoring methods will be better than the maximum likelihood estimation (MLE) if the prior distribution can be justified. Increasing the prior standard deviation from 1 to 2 can help overcome some of the regression problem in Bayesian estimations. When there is little information at the ability levels, expected a posterior (EAP) estimates have some advantages over maximum a posterior (MAP) estimates. At the ability levels where information is more adequate, the differences between the scoring methods decreased and MAP showed some advantages over EAP. The random entry level showed reasonably good results compared with the mean entry level. The results of study II (Biology test, 315 items) confirmed most of the predictions that study I made. When the information of the item pool increased, the differences among scoring methods decreased. If the ideal item pool is possible, the decision for the scoring method may not be as crucial. In the real world, EAP with prior SD of 2 showed the best result especially for the best result especially for the ability levels which do not have enough information.