A method to inspect the rattle generated in a vehicle cabin has been developed. In the method, the waveform of overall in-cabin noise is analyzed using Wigner distribution, a kind of time-frequency analysis, and the rattle component of the waveform is condensed and separated from the background shake noise. Then the rattle component is classified into three levels, strong, middle and not detected, using a neural network. Fuzzy inference is also used to select normal waveform data. Experimental results show that the correct classification ratio of the method is more than 90%, which equals the ability of skilled inspectors.