This research focuses on the efficiency assessment of Angolan thermal power plants using a Slacks Based Model with undesirable (bad) outputs (SBM-Undesirable). In this research, SBM-Undesirable is used first in a two-stage approach to assess the relative efficiency for these power plants. The most frequent cost structure variables adopted by literature to characterize a productive technology are used. In the second stage, Beta Regression models are combined with SBM-Undesirable efficiency scores to produce a model for predicting energy production performance. Results indicate that thermal power plants in Angola that are intensive in capital to the detriment of labor and that present lower unit capacity costs are more efficient and less polluting. Results also indicate the absence of a learning curve for increasing efficiency in energy generation. Policy implications are derived. (C) 2017 Elsevier Ltd. All rights reserved.