The enhancement of efficiency is world-wide trend to survive under intense competition. In recent years, the efficiency in the power industry is also one of the important topics. In case of nuclear power plants(NPPs), the period and quality of maintenance is an especially important factor to increase efficiency as well as availability. Therefore, the accurate identification of the root causes for lost electric output is indispensable to decrease the period and to increase quality of maintenance. The diagnosis in NPPs is more difficult because the turbine cycle of NPPs uses saturated steam as working fluid. In this study, authors tried to develop an advisory system with the quantitative diagnosis model consisting of statistical regression analysis and Bayesian network for the support of nuclear turbine cycle diagnosis. The proposed advisory system includes the knowledge-base representing the normal or abnormal behavior of nuclear turbine cycle. Authors selected 34 boundary parameters that independently influence to electric output. Using the data collected from a turbine cycle simulation tool, the statistical correlation between a boundary parameter and electric output was analyzed. To give the belief, that is the degree of accuracy, of root causes under various uncertainty sources, belief module for the boundary parameters is developed on the basis of Bayesian network. In conclusion, this diagnosis module can give the impacts of the root causes and their uncertainty simultaneously, so we call it 'Lost MW calculator'. After the validation using simulated data and actual performance data, this module was installed in Younggwang NPP units 3 and 4 in Korea. (c) 2005 Elsevier Ltd. All rights reserved.