To identify internal combustion engine (IC engine) noise, a mathematical model is established, and the corresponding program is developed. By introducing global thresholds to decrease noise in the wavelet transforms program, the anti-interference of the program can be guaranteed, the spectrums of the combustion noise and mechanical noise are separated, and their A-weighted sound power levels are figured out. Taking a CG125 style motorcycle engine for an example, theoretical examples that include different interferential noises are constructed to simulate the identification and to validate the accuracy and practicability of the program. Results of the identification show that the spectrums of the combustion noise and mechanical noise before and after separation are agreed well; errors of A-weighted sound power level are all less than 0.6 dB.