We have developed a fast and practical wave-equation-based migration method to image subsurface diffractors. The method is composed of three steps in our implementation. First, it decomposes extrapolated receiver wavefields at every imaging point into local plane waves by a linear Radon transform; the transform is realized by a novel computationally efficient recursive algorithm. Second, the decomposed plane waves are zero lag-correlated with the incident source wavefields, where the incident angles are computed via the structure tensor approach. The resulting prestack images are binned into dip-angle gathers according to the directions of the decomposed plane waves and the calculated incident angles. Third, a windowed median filter is applied to the dip-angle gathers to suppress the focused reflection energy, and it produces the desired diffraction images. This method is tested on synthetic and field data. The results demonstrate that it is resistant to random noise, computationally efficient, and applicable to field data in practice. The results also indicate that the diffraction images are able to provide important discontinuous geologic features, such as scattering and faulting zones, and thus are helpful for seismic interpretation.