The gunshot blast is one of the evidences used to identify the criminal gun in the shooting scene. We proposed a method of using the gunshot blast to identify the firearm caliber in terms of audio forensics. The experiments were performed on the database consisting of 80 audio files of gunshot sounds fired from 8 different guns. At first, the autocorrelation functions of the gunshot sounds were calculated and normalized in the preprocessing stage. Then the Fourier cosine coefficients could be used to reduce the fluctuation and to extract the features of gunshot sounds. In addition, the gun sounds could also be verified by evaluating the difference among the Fourier cosine coefficients of the reference, and then the gun caliber could be identified.