Recurrence characteristics of great circum-Pacific earthquakes and determinations of spatial distribution of seismic moment release are surveyed in order to delineate their general features and relate earthquake slip distribution to models of recurrent rupture. As noted by others, the pattern of moment release is typically very irregular, with strong concentrations in a few isolated regions of a much larger aftershock zone. Despite this complexity, rupture nucleation is notably systematic, with mainshock epicenter showing a strong tendency to locate in or immediately adjacent to identified regions of high moment release. This generalization suggests that earthquake recurrence is more likely to be controlled by maximum rather than average fault slip. Well-documented case histories from twelve plate boundary regions demonstrate that seismic strain release tends to be temporally well-ordered, while source dimensions, slip, and cumulative moment release vary considerably from cycle to cycle. On none of the segments studied do earthquakes consistently recur in nearly identical events. Instead, individual great earthquakes differ significantly from cycle to cycle or rerupture takes place in a sequence of two or more smaller events. Despite these differences the duration of the seismic cycle is approximately uniform. Furthermore, when strain release occurs in a sequence of large earthquakes, these events take place towards the end of the cycle and occur with increasing rupture length and magnitude. These ordered and irregular features of earthquake recurrence argue for the existence of corresponding elements on plate boundary faults. The ordered characteristics are identified with zones of concentrated moment release that slip comparable amounts in each cycle and have high shear strength. The irregular features are associated with intervening weaker regions that move in response to the stress concentration of dynamic rupture and slip by differing amounts in each event. Although these mechanistic associations are indirect and tentative, the features of recurrent behavior documented here have implications for long-term earthquake hazard assessment that are not dependent on the models proposed to explain them. Chief among these are the cycle-to-cycle differences among gap-filling events and the absence of shocks that fill major slip-deficient regions early in the seismic cycle.