Predicting electricity generation is important to developing forecasts of air pollutant release and to evaluating the effectiveness of alternative policies which may reduce pollution. A neural network model (NUMOD) predicting electricity generation fueled by coal, natural gas, and oil (whose combustion released air pollutants) was developed to run on a personal computer. NUMOD uses three linked, feed-forward neural networks, each trained with the extended delta-bar-delta paradigm. One network predicts coal-fired generation. Its output is fed as input to each of the other two networks: one for gas-fired generation and the other for oil-fired generation. In addition, all three networks use inputs describing electricity demand, fuel prices, generating equipment, climate, and power pooling. Pearson's r calculated at various points during training, out-of-sample tests, and performance evaluation was greater than 0.93 and frequently greater than 0.98.