A hybrid artificial neural network is used to model the thermodynamic behavior of the Tennessee Valley Authority’s Sequoyah nuclear power plant using data for heat rate measurements acquired over a 1-yr period. The modeling process involves the use of a selforganizing network to rearrange the original data into several classes by clustering. Then, the centroids of these clusters are used as the training patterns for an artificial neural network that utilizes backpropagation training to adjust the weights on the connections between artificial neurons. This procedure greatly reduces the training time and reduces the system error. Comparison of the calculated heat rates with those predicted by the artificial neural network gives an error of <0.1%. A sensitivity analysis is then performed by taking the partial derivative of the heat rate with respect to each individual input to secure a sensitivity coefficient. These coefficients identified the input variables that were most important to improving the heat rate and efficiency. The methodology reported is an alternative to the conventional modeling procedures used in other heat rate monitoring systems. It has the advantage that the artificial neural network model is based on actual plant data that cover the dynamic range normally occurring over an annual cycle of operation, and it is not subject to linearization or empirical approximations. This process could be utilized by existing heat rate monitoring systems.