Paleoseismic and historical earthquake records used to quantify earthquake recurrence rates can also be used to test the likelihood of seismically quiescent periods. At principal paleoseismic sites in California on the San Andreas, San Jacinto, Elsinore, and Hayward faults, no ground‐rupturing earthquake has occurred in the last 100 yr, yet this interval is about three times the average interearthquake period for the ensemble of sites. We examine long paleoseismic records from these faults, as they carry most of the transform fault slip on the plate boundary, to see if the current hiatus has any precedent in the last 1000 yr. The selection of sites is designed to sample fault sections unlikely to have ruptured together, so their conditional probabilities of a hiatus can be combined as independent events. We find a 100‐yr hiatus is not predicted by common time‐dependent or time‐independent recurrence models. Paleoearthquake dating uncertainties can allow long open intervals at individual sites or subsets of sites, but do not explain the observed gap in the ensemble. After approximately removing redundancies in the full paleoearthquake record, the time‐independent probability of the current 100‐yr gap is of order 0.3%. This raises several questions. Do we live in a statistically exceptional time? Or does some wide‐scale effect modulate earthquake occurrence among sites over longer timescales? Finally, how should we understand seismic hazard estimates in California if the recurrence models on which they rely seem, at minimum, incomplete? Whether due to a statistical anomaly, some longer‐term modulation of earthquake occurrence, or another cause, our results emphasize that the hiatus of the last century has been exceptional.