This paper demonstrates how synthetic seismicity calculations which are based on the concept of fault segmentation and incorporate the physics of faulting through static dislocation theory can improve earthquake recurrence statistics and hone the probabilities of hazard. Compared to forecasts constructed from a handful of earthquake recurrence intervals, forecasts constructed from synthetic seismicity are more robust in that they embody regional seismicity information over several units of magnitude, they can extrapolate seismicity to higher magnitudes than have actually been observed, and they are formulated from a catalog which can be extended as long as needed to be statistically significant. Synthetic seismicity models can also be used to judge the stability of common rate estimates and the appropriateness of idealizations to the earthquake cycle. I find that estimates of fault slip rate are unbiased regardless of sampling duration, while estimates of earthquake recurrence time are strongly biased. Recurrence intervals estimated from seismicity samples less than about 10 times the actual recurrence interval will almost certainly be too short. For the Middle American Trench (MAT), it would take 200 and 400 years of monitoring to constrain slip and recurrence rates to ¿10%. Events M≥6 have as much as a 60% probability of recurrence within 5 years due to the clustering of small earthquakes in foreshocks and aftershocks. This probability drops to less than 15% for M≥7 events. Increasing gap time generally increases conditional probability of earthquake occurrence, but the effect is weak. For the MAT, the spread parameters of the best fitting lognormal or Weibull distributions (≈0.75) are much larger than the 0.21 intrinsic spread proposed in the Nishenko-Buland hypothesis. Stress interaction between fault segments disrupts time or slip predictability and causes earthquake recurrence to be far more aperiodic than has been suggested. ¿ American Geophysical Union 1992