In recent years, nuclear accidents have spurred the improvement of nuclear power plants, because nuclear pollution has resulted in considerable environmental damage. This paper describes a radial basis function (RBF) neural network prediction model based on a genetic algorithm to predict concentrations of nuclear leakage in case of nuclear accidents. The prediction model uses various factors impacting the concentrations of nuclear leakage as input and concentrations of nuclear leakage concentrations as output. We used data from the Hong Yanhe Nuclear Power plant in the Liaoning Province for a simulation experiment, and compared the proposed prediction model with a RBF prediction model based on a traditional genetic algorithm. The results demonstrated that the prediction model had a smaller error than the output from the traditional model.