Maximum earthquake magnitude and the rate of seismic activity apparently differ among subduction zones. This variation is attributed to factors such as subduction zone temperature and stress, and the type of material being subducted 1-5 . The relative velocity between the downgoing and overriding plates controls their tectonic deformation. It is also thought to correlate with seismicity 1,2,6-8 . Here I use the epidemic type aftershock sequence model 9,10 to calculate the background seismicity rate-the frequency of seismic events above magnitude 4.5-for 117 sections of subduction zones worldwide, during the past century. I demonstrate a proportionality relationship whereby relative plate velocity correlates positively with seismicity rate. This relationship is prominent in the southwestern Pacific Ocean. However, although seismically active, this region has not experienced a magnitude 9 earthquake since 1900. In contrast, the Cascadia, Nankai, southern Chilean and Alaskan subduction zones exhibit low background seismicity rates, yet have experienced magnitude 9 earthquakes in the past century. Slow slip occurs in many of these regions, implying that slow deformation may aid nucleation of very large earthquakes. The proportionality relationship could be used to assess the seismic risk between two endmembers: active subduction zones that generate moderate earthquakes and quiet subduction zones that generate extremely large earthquakes.It has long been discussed whether the characteristics of seismicity are spatially variable throughout the world's subduction zones. Pioneering studies 1,2 indicated that the largest earthquake possible for each subduction zone is determined mainly by the subduction velocity and the age of the subducting plate, exemplified by two endmembers: the fast and young Chilean subduction zone, and the old and slow Mariana subduction zone. Thirty years later, this simple concept requires revision 11-13 . For example, an alternative explanation for maximum earthquake size, based on orientation of the stress field and the size of the accretionary prism, has been proposed 4 .Although information on maximum earthquake size is directly useful in seismic risk assessment, it is difficult to estimate because the period over which instrumental observations have been made is shorter than the long recurrence time of very large earthquakes. The length of observation cannot be extended quickly and information on maximum earthquake size has not advanced significantly in the past 30 years, although improvements in seismic observation systems and data processing schemes have enabled detailed discussions on the statistical characteristics of moderate and large earthquakes. For example, the b-value of the Gutenberg-Richter relation shows a statistically significant difference among some subduction zones 13 and swarm-like activity may be spatially limited