Iran Khodro Company is a leading auto maker in Iran which holds about 65% of the market share and hence the share holders show great interests in predicting its stock exchange price. On the other hand due to the chaotic behavior of share price in Tehran Stock Exchange the classical models such as ARIMA and ARCH would not be efficient models to represent the dynamics governing the share price. However, neural network (NN) models are proposed to predict Iran Khodro Stock Exchange Price (IKSEP). Several neural network models such as MLP, ELMAN, CASCADE, GRNN and RBFN were examined. Because of serious volatility in IKSEP, special method was proposed for testing and training the data which considerably improved the results. Extensive tests have been curried out to choose the most suitable feature such as, the type of transfer function, the number of hidden and output layers, the training algorithm, and the technical and fundamental variables. Some fundamental variables such as oil price, P/E and volume of stock exchange were introduced in the model and showed to be considerably effective in the accuracy of forecast. The best results obtained from NN models were compared to those obtained by using exponential smoothing and Box-Jenkns models. The results showed the NN forecasts were superior to those of the time series model.