The characteristics of the last stage group of condensing steam turbine are analyzed, and a method based on the synthetic BP (back-propagation) neural network is proposed for determining the normal value of relative internal efficiency of the last stage group. In order to consider the influence of the regenerative system, the influential factors of the relative internal efficiency of the last stage group firstly are determined, and the corresponding mathematical model is set up, and finally using the BP neural network to fit the equation. In this paper, the relative internal efficiency could be calculated by the method of BP instead of finding the exhaust enthalpy in the wet region first. The results show that the average relative errors between the network output values and the calculated values of off-design condition are less than 1%, which verify the accuracy, feasibility and validity of this method. (C) 2016 Elsevier Ltd. All rights reserved.