本文探討實驗設計的全域最佳化問題，提出一使用田口直交表之實驗設計，發展出擇優型(P-OA)以及去劣型(E-OA)兩種迭代式搜尋法。 本文使用13個標竿測試函數，包含有單極值與多極值的函數。首先，將本文搜尋法與遺傳演算法(Genetic Algorithm)進行比較搜尋結果，由測試結果顯示，不論是單極值函數或多極值函數，本文所提之迭代式搜尋法的收斂速度遠大於遺傳演算法，而且搜尋到的最佳解也比遺傳演算法要好。 再者，由一系列的實驗設計問題測試結果得知：本文演算法依設計問題的維數，使用三水準中實驗組數最少的直交表，以處理多水準數的實驗設計問題，其中本文兩種不同型態的演算法對於等高線圖為圓形的函數都能夠有好的表現，但E-OA比P-OA的可靠度、穩定度及精確度為高；於工程及有限制條件式的問題中，本文兩種演算法也都有不錯的表現。 In this study, we discuss the design of experiment in global optimum problems, and propose an iterative approach with use Taguchi orthogonal array. The two iterative approaches Pickup-type (P-OA) and Eliminated-type (E-OA), are developed. We use 13 standard test functions including single extreme and multi-extreme functions in this study to study the performance of these two approaches. First, in this study we compare those results of these two approaches and Genetic Algorithm. The results showed that even if for single extreme or multi-extreme functions, the convergent speed of these two approaches in this study is quicker than Genetic Algorithms, and the found best solutions are more accurate than Genetic Algorithms. Second, a serial of design of experiments with several different dimensions/factors and levels has to be conducted. The results show the two approaches using the least number of trials in 3-level orthogonal array in this study are more suitable to the problem with a circle shape. The performance, of E-OA, such as reliability, stability, and accuracy, is better than those of the P-OA. The two approaches in this study also have good results in engineering and problems with limit conditions.