Empirical mode decomposition (EMD) is a new method to process nonlinea r and nonstationary signals. However, there is some trouble with its endeffect due to using spline interpolation to get its upper and lower envelopes. Based o n a brief discussion to the cause of the endeffect, the mirror extension and r adial basis function (RBF) neural network prediction are adopted to suppress the endeffect in EMD. A simulated signal and a signal recorded from the shaking table test of a 12stroey reinforced concrete frame model are processed by the two methods and further decomposed by EMD. The results indicate that the two met hods predict and extend the data according to the boundary information of the si gnal and can suppress the endeffect effectively, thus improving the decomposition r e sults to a certain extent. In addition, for a complex signal, such as a measured acceleration signal of a structure, only extending the signal by RBF neural net work is not effective in suppression of the endeffect. But applying mirror ext ension after filtering the signal and predicting the filtered signal by RBF neur al network can suppress the endeffect obviously.