To overcome the limitations of low frequency resolution and interference of aliasing distortion of neighbouring harmonic in the vibration signal analysis of diesel engine using the wavelet transform(WT)or the wavelet packet transform(WPT),a novel fault diagnosis method based on the empirical mode decomposition(EMD)and the support vector machine(SVM)is proposed.The method employs EMD to decompose the vibration signal into many intrinsic mode functions(IMFs),and extracts the wavelet packet features of several energy-dominating IMFs respectively.Then,the SVM-based classifiers are trained in each of featrre sub-spaces and fused by the energy weight of each IMF.The proposed metheod is applied to fault diagnosis of a 6135-type diesel engine,the results show that the independent IMF analysis for tault diagnosis is feasible,and the proposed method can effectively diagnose the conventional faults of 6135-type diesel engines.