Monetary policy makers have been engrossed continually to discover the suitable method for exchange rate forecasting in order to preventing its disruptive movements. But exchange rate movement's manifold determinants cause to its complex and nonlinear behavior. Nonlinear models have better performance for its forecasting. Therefore, in this study the performance of nonlinear models such as ANFIS and NNARX and linear model such as ARIMA were compared in Iran's Rial/US$ & Rial/€ for 2, 4 and 8 days ahead via most important forecast performance criteria and daily data related to 20 March 2002- 21 November 2008. Results indicated that ANFIS and NNARX models have better performance in comparison with ARIMA model, and ANFIS model outperforms NNARX model in all horizons of Iran's Rial/US$ & Rial/€ exchange rate forecasting.
