Following the Chalfant Valley earthquake of 1986 July 21 (ML= 6.4), the USGS deployed 120 vertical-component seismometers to obtain dense array recordings of aftershocks. From this data set, we selected three events with the best station coverage to obtain acoustic images of the earthquake sources. the events were relocated using a layered velocity model and models with a constant velocity and a constant gradient, to obtain minimum traveltime residuals. All models performed similarly well. Preprocessing of the data prior to the extrapolation consisted of applying static corrections, lowpass filtering and spatial interpolation as well as spatial tapering at the edges of the recording line. For the back projection of the recorded wavefield, we used a second-order finite-difference reverse-time extrapolation technique similar to McMechan (1982). While for two events, no focusing of the energy could be obtained, an acoustic image of one earthquake source was retrieved. In order to obtain a better understanding of the physical meaning of acoustic imaging, we applied the same processing and extrapolation scheme to synthetic seismograms calculated using the reflectivity method. For the present data set these tests indicate that the influence of the 1-D velocity model on the focusing quality is very low. the most severe limitations, however, are due to the finite aperture and gaps of the recording line. the spatial sampling of the wavefield limits the frequency range used for extrapolation, so that the resolution is beyond the scale of the source processes. the finite aperture leads to polarity reversals in the final source image, which makes an interpretation of the radiation pattern at the time of focusing from the reconstructed image impossible.