Adjacent segments of the plate boundary may fail separately as earthquakes that occur very close in time. When the time between events is a small fraction of the recurrence interval, then the sequence is referred to as ''doublets'' or ''multiplets''. In the first part of this paper, we determine the frequency of occurrence of large multiplet earthquakes. We count the number of multiplet earthquakes present in the Abe catalogue with aftershocks removed (Ms≥7.1) using a series of space, time, and magnitude filters. We compare this number to the number of multiplet earthquakes present in the Abe catalogue where origin times of earthquakes are randomized and epicentral locations and magnitudes are kept constant. The difference between the observed and random number of multiplets gives us the percentage of multiplets that do not occur randomly. We find that this percentage ranges from 2% to 7% of all large earthquakes based on the best choice of space, time, and magnitude filters. In the second part of this paper, we model multiplet earthquakes using three simple frictional slider models. The models consist of slider blocks resting on a frictional ''conveyer belt'' and interconnected by springs. The seismogenic portion of the plate interface is represented by blocks that exhibit stick-slip behavior (asperities). Aseismic creep within the seismogenic zone is represented by ''creepers''. The ''creepers'' are blocks that obey a simple linear creep law. The three models differ in that the positions of the creeping blocks within the system of asperities and creepers are different. We find that model 3 produces a complex sequence of events consisting of a single, double, and multiplet earthquakes. These synthetic sequences of events are comparable to the earthquake cycle observed along Nankai trough. Model 3 produces a high percentage of multiplet earthquakes when the creeper is allowed to relax within a small fraction of the maximum recurrence time of the event sequence. This implies that the details of the frictional law invoked for the creeper blocks does not influence the model results significantly. The time evolution of model 3 is compared to the postseismic phase of the loading cycle at Muroto Point, Japan. We suggest that the rapid postseismic deformation that occurs immediately after the Nankaido earthquake is due to a rapid creeping zone that is in this region. ¿ American Geophysical Union 1996