本研究發展新的計量方法，運用隨機邊界法將單調性和曲度條件等性質納入迴歸模型，使用聯立迴歸模型進行估計，藉此讓係數估計值不易出現違反經濟理論的情況。 採用台灣2002年至2015年，51家商業銀行進行實證分析，發現本研究方法估計產出方向距離函數時，僅有5%以下的樣本點不符合單調和曲度等性質，用於估計產出面距離函數時，僅有2%以下的樣本點不符合。進一步探討台灣銀行業之非意欲產出--逾期放款--對估計技術效率的影響，顯示不考慮此非意欲產出造成整體銀行業、非金控本國銀行與外商銀行的技術效率被高估，而金控本國銀行的技術效率則被低估，此外，分析2007年金融風暴前後銀行業經營效率變化，顯示考慮非意欲產出銀行業經營效率顯著提升，然而，未考慮非意欲產出銀行業經營效率卻下降，兩者結果有著極大的差異。 / The aim of the paper is to develop a new approach, which is stochastic frontier analysis imposing monotonicity and curvature conditions, then using simultaneous regression model to estimate. By the approach, it can solve the problem of most of the coefficient estimates violating the economic theory. The study uses the data of 51 commercial banks in Taiwan from 2002 to 2015 to conduct the empirical analysis. It indicates that by output directional distance function, less than 5% sample points violate the monotonicity and curvature conditions; by output distance function, less than 2% sample points don’t obey the restricted conditions. Further, the paper discusses the effect of commercial banks’ undesirable output- non-performing loan- on estimating technical efficiency. The results show that ignoring the undesirable output cause the technical efficiency of overall banks, non-finance holding banks and foreign banks are overvalued, and the technical efficiency of finance holding banks are undervalued. Furthermore, analyze the change of business efficiency after financial crisis in 2007. It points out that using the model consider the undesirable output, the banks’ efficiency rises. However, using the model no consider the undesirable output, the banks’ efficiency decreases. There is an extremely conflict between two approach.