This study applied back-propagation neural network (BPNN) and empirical mode decomposition (EMD) techniques for forecasting exchange rate. The aim of this study is to examine the feasibility of the proposed EMD-BPNN model in exchange rate forecasting. In the first stage, the original exchange rate series were first decomposed into a finite, and often small, number of intrinsic mode functions (IMFs). In the second stage, kernel predictors such as BPNN are constructed for forecasting. It was demonstrated that the proposed model performs better than traditional model (random walk). The mean absolute percentage errors are significantly reduced.