本文探討在一元配置 (one-way layout) 設計之下的劑量反應研究中,鑑別藥物最低有效劑量(minimum effective dose,簡記MED),此處的最低有效劑量為相對於零劑量組或標準組,具有一定藥效的劑量水準。本文考慮劑量反應為常態分布或是非常態分布,並且各劑量組分散度不盡相同時,在設定的藥效門檻值之下,提出一種修正的Mann-Whitney (1947) 統計量,建立封閉性的多重檢定方法進行藥物最低有效劑量之無母數鑑別。然後利用蒙地卡羅 (Monte Carlo) 方法在各種可能的劑量反應之下比較本文所提檢定方法與其他方法之實驗誤差率 (experimentwise error rate,簡記EWE)、族誤差率(familywise error rate,簡記FWE)及檢定力 (power) 之相對表現。最後進行實例分析說明所提方法之應用。 This paper discusses in one-way layout dosage response research, how to identify minimun effective dose, denote it bu MED. Corresponding to the dose response with normal or nonnormal distribution, maybe in different scale parameter, and with threshold set-up, we propose a modified Mann-Whitney statistics, set up a close and step-down multiple comaprison to identify nonparamerically MED.