A 24-hour continuous sound recording was made after the 1986 May 20 earthquake. During the 24 hour period, 25 aftershocks with a magnitude greater than 2 Were recorded by a temporary seismographic network. Comparing the seismograms with the recorded sounds, we found that 5 of these aftershocks were accompanied by earthquake sounds, and all of these sounds preceded the P-wave arrivals by 0.2 to 1.1 seconds. These 5 aftershocks possessed common characteristics of greater magnitude, shallower focus, and being closer to the recording site than other aftershocks. The predominant frequencies of the earthquake sounds ranged from 25 to 100 Hz. These predominant frequencies were relatively lower when the focus was deeper. Based on these discoveries, we infer that, before the main faulting which resulted in the earthquakes, there were associated cracks that sent out acoustic waves with frequencies higher than those detected by the seismographs.