We deployed a dense linear array of 45 seismometers across and along the San Andreas fault near Parkfield a week after the M 6.0 Parkfield earthquake on 28 September 2004 to record fault-zone seismic waves generated by aftershocks and explosions. Seismic stations and explosions were co-sited with our previous exper- iment conducted in 2002. The data from repeated shots detonated in the fall of 2002 and 3 months after the 2004 M 6.0 mainshock show 1.0%-1.5% decreases in seismic-wave velocity within an 200-m-wide zone along the fault strike and smaller changes (0.2%-0.5%) beyond this zone, most likely due to the coseismic damage of rocks during dynamic rupture in the 2004 M 6.0 earthquake. The width of the damage zone characterized by larger velocity changes is consistent with the low-velocity waveguide model on the San Andreas fault, near Parkfield, that we derived from fault-zone trapped waves (Li et al., 2004). The damage zone is not symmetric but extends farther on the southwest side of the main fault trace. Waveform cross- correlations for repeated aftershocks in 21 clusters, with a total of 130 events, located at different depths and distances from the array site show 0.7%-1.1% increases in S-wave velocity within the fault zone in 3 months starting a week after the earthquake. The velocity recovery indicates that the damaged rock has been healing and regaining the strength through rigidity recovery with time, most likely due to the closure of cracks opened during the mainshock. We estimate that the net decrease in seismic velocities within the fault zone was at least 2.5%, caused by the 2004 M 6.0 Parkfield earthquake. The healing rate was largest in the earlier stage of the postmainshock healing process. The magnitude of fault healing varies along the rupture zone, being slightly larger for the healing beneath Middle Mountain, correlating well with an area of large mapped slip. The fault healing is most promi- nent at depths above 7 km.