本研究應用模擬降溫（simulated annealing）方法於自動化編製平行測驗（the automated construction of parallel tests）。模擬降溫方法可對combinatorial optimization problems提供優良的解決答案，而且在運算與時間上都相當經濟有效。本研究使用的題庫有二，分別是96題的圖形題庫與84題的語文題庫。研究目標在於以模擬降溫方法，從每一題庫中各自編製一組30題的平行測驗。本研究設置以及比較兩種選題方式：組間配對（pairwise selection）與測驗配對（group selection）。主要研研結果發現，測驗配對方式編製出來的平行測驗比組間配對方式更為平行，且其information也較多。至於使用組間配對方式的四種方法中，也有兩項結果：(1)配對item information所編製的平行測驗比配對試題參數估計值（item parameter estimates）更為平行，(2)不等比重的配對試題參數估計值（differential weighting of item parameter estimates）也許會比相等比重的配對試題參數估計值（equal weighting of item parameter estimates）更能提高測驗的平行程度。

　　　　　The construction of parallel tests was automated with the application of simulated annealing in the study. The method of simulated annealing provided good solutions to the combinatorial optimization problems involved, and was efficient in the computational efforts and time required. The task in the study was to construct two 30-item parallel tests form each of the item banks: the 85-item figural item bank and the 95-item verbal item bank. Two approaches of item selection were implemented and compared: pairwise selection and group selection. it is found that matching the global similarity of tests (group selection) mode better parallel tests with higher information than matching the pairwise similarity of items (pairwise selection). there were also two findings related to pairwise selection: (1)Matching item information for the pairs made better parallel tests than matching item parameter estimates, and (2)differential weighting of the item parameter estimates may increase test parallelism as opposed to equal weighting.