Abstract: For the moving target detection using the through wall radar, according to the fact that the Doppler signals of the human activities are nonlinear and non-stationary, the ensemble empirical mode decomposition (EEMD) and empirical mode decomposition (EMD) are used respectively to decompose five different activities Doppler signals into a set of instinct mode functions (IMF). The five different activities include the human standing still, standing with arms waving, stepping forward and backward, walking and running. A support vector machine (SVM) is trained using the energy ratio of each IMF to the total IMFs as the features to classify the activities. The relationship between the classification accuracy with the number of the features is analyzed and the classification accuracy comparison using two different decomposition methods is given. Because the EEMD can eliminate the mode mixing problem existed in EMD and each IMF obtained from the EEMD has a clear physical meaning, the classification accuracy using the EEMD is found to be more than 94%, higher than the one using the EMD.