An automatic failure diagnosis system for engines where we apply motoring test bed and sampled firing test set is proposed in this paper. Focusing on the inspection items where sound signals can be employed, we propose a new approach utilizing wavelet transform, FFT analysis and cepstrum analysis to anomaly characterization. For three important items, namely tappet clicks, misfire and mismatching, the following results have been obtained. (1) For tappet clicks, motoring test is more effective than firing test in recognizing the features of the anomaly. Wavelet transform is more reliable and convenient than FFT analysis to extract the features. (2) For misfire, cepstrum analysis is simple and effective to detect the features. (3) For abnormal combustion sound due to time mismatching, FFT analysis can be applied to extract the features. Further studies are required for finding more effective methods.