A new method of fault diagnosis of diesel engine based on HHT marginal spectrum is put forward in this paper. A lot of diesel engine faults were simulated and investigated on the diesel engine type 3110, and surface vib ratio n signals were obtained under both fault and normal conditions. Interval analysis method was used to analysis the property of the signals in th e time domain. The H HT(HilbertHuang transformation) marginal spectrum method was used to analyze t he segments of the signal which are sensitive to the faults, and an exact expre ssion of the variatioin of the signal with time and frequency was obtained. The maximum peak value of the marginal s pectrum was adopted as the feature vector and served as the input param eter of Ma halanobis distance classifier to classify working condition of the diesel engine . The results show that the proposed approach can classify working condition of the diesel engine accurately and effectively even under conditons of a small sample size.