濁水溪不僅為台灣第一長河川，其暴漲猛落的水文特性亦為西部河川的典型，多年來上游集水區變遷與颱風、豪雨於該區造成多次災害。為減低颱風豪雨可能在濁水溪流域帶來之洪水災情，因此有必要建立洪水預警模式，對颱洪事件進行有效之預測，以降低洪水所帶來之生命財產損失。反傳遞模糊類神經網路為一包含輸入層、隱藏層及輸出層的模糊類神經網路，其主要以規則庫控制為基礎，結合模糊控制及反傳遞類神經網路。本研究將此一模式架構於濁水溪流域，對暴雨時期逕流量進行相關性之評估，並應用於未來逕流量之預測。經驗證，可得到相當良好之結果。 The Choshui River inherited severe changes in stream flow regime with time is the longest river in Taiwan. Natural disasters such as floods, typhoon, and debris flows have been encountered and caused a great amount of damages and the changes of the hydrological characteristics of the basin. Owing to this reason, building a forecast model to reduce the flooded damage of the Choshui River is important and necessary. The structure of the counterpropagation fuzzy neural network is a kind of fuzzy neural network, which represented as an input layer, a hidden layer, and an output layer. This neural network is constructed by a set of rule-base control, fuzzy control, and counterpropagation network. This research presents a CFNN approach to the estimation of stream flow of the Choshui River in Taiwan. The results demonstrate that the ability of the approach is superior in terms of high prediction accuracy.