The concept of the earthquake cycle is so well established that one often hears statements in the popular media like, "the Big One is overdue" and "the longer it waits, the bigger it will be." Surprisingly, data to critically test the variability in recurrence intervals, rupture displacements, and relationships between the two are almost nonexistent. To generate a long series of earthquake intervals and offsets, we have conducted paleoseismic investigations across the San Andreas fault near the town of Wrightwood, California, excavating 45 trenches over 18 years, and can now provide some answers to basic questions about recurrence behavior of large earthquakes. To date, we have characterized at least 30 prehistoric earthquakes in a 6000-yr-long record, complete for the past 1500 yr and for the interval 3000-1500 B.C. For the past 1500 yr, the mean recurrence interval is 105 yr (31-165 yr for individual intervals) and the mean slip is 3.2 m (0.7-7 m per event). The series is slightly more ordered than random and has a notable cluster of events, during which strain was released at 3 times the long-term average rate. Slip associated with an earthquake is not well predicted by the interval preceding it, and only the largest two earthquakes appear to affect the time interval to the next earthquake. Generally, short intervals tend to coincide with large displacements and long intervals with small displacements. The most significant correlation we find is that earthquakes are more frequent following periods of net strain accumulation spanning multiple seismic cycles. The extent of paleoearthquake ruptures may be inferred by correlating event ages between different sites along the San Andreas fault. Wrightwood and other nearby sites experience rupture that could be attributed to overlap of relatively independent segments that each behave in a more regular manner. However, the data are equally consistent with a model in which the irregular behavior seen at Wrightwood typifies the entire southern San Andreas fault; more long event series will be required to definitively outline prehistoric rupture extents.