One of the traditional methods for forecasting is time series analyzing, which based on tow assumption: stationary and linearity. There is Suspicion to working by the traditional models. One of the substitute methods is Artificial Neural Network (ANN) which sometimes shows good potential ability for forecasting time series. In this paper, after reviewing the recent studies on ability of Autoregressive Integrated Moving Average Process (ARIMA) and ANN, tow models for forecasting the exchange rate between march 2006 and February 2009 in Iran are compared. The results show that ANN has a better estimate than ARIMA. In this research, the MATLAB and Central bank data are utilized.