This paper employs a stochastic frontier analysis (SFA) to examine cost efficiency and scale economies in Taiwan Power Company (TPC) by using the panel data covering the period of 1995-2006. In most previous studies, the efficiency estimated by the Panel Data without testing the endogeneity may bring about a biased estimator resulting from the correlation between input and individual effect. A Hausman test is conducted in this paper to examine the endogeneity of input variables and thus an appropriate model is selected based on the test result. This study finds that the power generation executes an increasing return to scale across all the power plants based on the pooled data. We also use installed capacity, service years of the power plant, and type of fuel as explanatory variable for accounting for the estimated cost efficiency of each plant by a logistic regression model to examine the factor affecting the individual efficiency estimates. The results demonstrate that the variable of installed capacity keeps a positive relationship with cost efficiency while the factor of working years has a negative relationship. Crown Copyright (C) 2009 Published by Elsevier Ltd. All rights reserved.