On 28 September 2004 an M 6.0 earthquake occurred, rupturing a 30-km stretch of the San Andreas fault near Parkfield, California. The fault segment, which hosted a cluster of nearly identical microearthquakes (a multiplet), had been under close surveillance since the mid-1980s following a long-term prediction by the United States Geological Survey that a characteristic earthquake of M∼6.0 would occur near Parkfield before 1993. The multiplet made possible a study of temporal changes in seismic propagation characteristics associated with the mainshock. We show that a sharp rise in P-wave attenuation operator t* began to appear about 18 months before the belated seismic event, accompanied by a rapid fall in P/S ratio, in both the frequency and time domains. The P waves that interrogated the northern half of the impending rupture zone, where the coseismic slip was to be the largest, yielded the strongest t* anomalies. The anomalies peaked six weeks after the Parkfield earthquake and then diminished rapidly in the months ensuing. The preseismic rise of the t*, accompanied by a rapidly falling P/S ratio, pointed to a decline in pore-fluid saturation due to microcracking that culminated in the catastrophic rock failure during the fault rupture.