In this paper, we compare the properties of stochastic process with deterministic chaos system using the AR(l) model as the comparing model. 126 neural networks are built to approximate chaotic system for the 18 data sets generated by 6 AR( 1) models. From predicting capabilities of the neural networks, we can appropriately deside the dimensions of the chaotic systems. Also, the emperical results show that the forecasting performance derived from neural networks is more robustic than those derived from the AR( 1) model.