With its high repeatability，air-gun source has been used to monitor the temporal variations of subsurface structures. However，under different working conditions，there will be subtle differences of the air-gun source signals. To some extent，deconvolution can eliminate changes of the recorded signals due to source variations. Generally speaking，in order to remove the air-gun source wavelet signal and obtain the Green's functions between the air-gun source and stations，we need to select an appropriate method to perform the deconvolution process for seismic waveform data. Frequency domain water level deconvolution and time domain iterative deconvolution are two kinds of deconvolution methods widely used in the field of receiver functions，etc. We use the Binchuan(in Yunnan，China)air-gun data as an example to compare the performance of these two deconvolution methods in air-gun source data processing. The results indicate that frequency domain water level deconvolution is better in terms of computational efficiency；time domain iterative deconvolution is better in terms of the signal to noise ratio(SNR)，and the initial motion of P wave is also clearer. Besides，we further discuss the sequence issue of deconvolution and stack for multiple-shot air-gun data processing. Finally，we propose a general processing flow for the air-gun source data to extract the Green's functions between the air-gun source and stations.