Abstract: The seismic response characteristics of collapsed columns are usually expressed in the form of diffraction， so diffractions are effective wave fields to identify and track collapsed columns. In order to effectively use the diffrac⁃ tion to accurately predict the collapse column in coal field， a method for separating seismic diffraction via median stop filter was developed based on the median filter theory and the idea of frequency domain band stop filter， and an effec⁃ tive diffraction imaging processing flow was established. 1 The key to the utilization of diffraction is how to completely separate diffraction waves. Due to the difference of time distance characteristics between diffraction and reflection， the median stop filter based on statistical characteristics has the possibility of separating diffraction and re⁃ flection. 2 To effectively suppress the reflection， it is necessary to increase the transverse coherence difference be⁃ tween reflection and diffraction through NMO， and then implement median stop filtering to separate the diffraction for the one⁃dimensional seismic signal with the same sampling time on different seismic channels. 3 To ensure the complete extraction of diffraction， the window factor of the median stop filter is determined quantitatively by establishing the relationship between the main wavenumber band and the number of seismic channels. 4 The separa⁃ ted diffraction and full wave field are imaged by pre⁃stack seismic migration imaging method， and then the coalfield collapse column will be predicted by combining the diffraction and full wave field imaging results. The cor⁃ rectness of the method was verified by a typical coalfield collapse column model， and it was further applied to the pre⁃ diction of collapse column in Shanxi Z coalfield. The results show that the diffraction migration imaging results are ob⁃ viously better than that of the traditional reflection. The diffraction migration imaging can clearly reveal the edge pro⁃ file， scale and other information of the collapse column， and can effectively improve the identification accuracy of the coalfield collapse column.