We have compared digitally-recorded waveforms of ML 2.0--2.8 earthquakes that occurred in two small areas along the Imperial fault before and after it broke in the ML 6.6 Imperial Valley earthquake on October 15, 1979. Eight preshocks (1977--1979) from a 4 1/2 by 1 1/2 km area centered 4 km SE of the mainshock epicenter have strikingly similar waveforms over the entire record length (~30 s), with an average peak cross correlation between seismograms of 0.74. The seismograms are well correlated at frequencies up to at least 4 Hz. This implies similar source mechanisms and hypocenters with 1/4 of the 4-Hz wavelengths, i.e., <200--400 m. Five aftershocks from the same area show an average peak cross correlation between seismograms of only 0.23. Any associated changes in mechanism must be small because they are not reflected in the first motion data. Analysis of frequency content of these events using bandpass-filtering techniques showed no systematic temporal changes in spectral shape. Ten preshocks and 24 aftershocks from a 1 1/2 by 2 km source area centered along the fault 16 km NW of the mainshock epicenter were also studied. First motion data suggest that all of the aftershocks and a swarm of six preshocks on December 7--9, 1978, were associated with the main fault but that four earlier preshocks were not. The six preshocks on December 7--9, 1978, were tightly clustered, as evidenced by the strong similarity of the waveforms (most peak cross correlations ≥0.6). During this swarm the 8- to 16-Hz spectral amplitude increased relative to the 1- to 2-Hz spectral amplitude over the whole record length by about a factor of 3, suggesting a systematic increase in stress drop. Groups of like events are also present among the aftershocks in this data set. The average peak correlation for pairs of aftershocks, 0.43, is almost the same as that for pairs of preshocks, 0.45, if all 10 preshocks are included. However, several sources appear to have been active simultaneously during the aftershock period so that no more than two to three consecutive aftershocks have maximum cross correlations ≥0.6. The highly localized sources characterized by waveform similarity may represent fault asperities or clusters of asperities. Our observations are consistent with a decrease in the number of these asperities as the weaker ones fail under increasing stress during the intervals between large earthquakes.