Correlations among subduction zone seismicity, convergence rate and subducting plate age are reassessed considering the possible roles of both recurrence times and fault zone temperatures. Distributions of earthquakes with respect to subducting lithosphere age and convergence rate are grossly explained by a recurrence relation when ages and rates at the world's trenches are taken into account. Correlations between maximum earthquake size Mwmax and convergence rate occur because faster subduction lowers the average recurrence time, so that at random within a limited sampling time, faster subduction zones have larger earthquakes. Published empirical slopes of an assumed linear relationship between Mwmax and convergence rate are predicted to within 1 standard deviation by such a recurrence model. Recurrence predicts that Mwmax should be related to the logarithm of convergence rate and revised age-rate-Mwmax data agree with such a relationship. No resolvable global correlation between age and Mwmax is found. Hence mechanical explanations of subduction zone seismicity based on such correlations are not required. Predicted average steady state temperatures, based on age and dip of the subducting lithosphere and convergence rate, at most subduction zone thrust faults are within a small range of values (¿50 ¿C). The few warm outliers, that is, Cascadia, Mexico, and southernmost Chile (south of Chile Rise) that subduct very young lithosphere, also have low seismic coupling coefficients suggesting that high temperature may inhibit seismicity. Fault zone temperatures can explain global variations in seismic coupling coefficients as well as the slab anchor model of Scholz and Campos <1995>. Applying this to the Cascadia subduction zone, in contrast to conclusions drawn from mechanical models of subduction and consistent with many other observations, due to its extremely high temperature, Cascadia may be a region where aseismic subduction predominates.¿ 1997 American Geophysical Union