In this paper we use neural network models to forecast earnings per share (EPS) of Chinese listed companies using fundamental accounting variables. The sample includes 723 Chinese companies in 22 industries over 10 years. The result shows that the neural network model with weights estimated with genetic algorithm (GA) outperforms the neural network with weights estimated with back propagation (BP). Results also show that the addition of fundamental accounting variables used in the neural network models further improves the forecasting accuracy.