Rates of foreshock occurrence involving shallow M ??? 6 and M ??? 7 mainshocks and M ??? 5 foreshocks were measured in two worldwide catalogs over ???20-year intervals. The overall rates observed are similar to ones measured in previous worldwide and regional studies when they are normalized for the ranges of magnitude difference they each span. The observed worldwide rates were compared to a generic model of earthquake clustering based on patterns of small and moderate aftershocks in California. The aftershock model was extended to the case of moderate foreshocks preceding large mainshocks. Overall, the observed worldwide foreshock rates exceed the extended California generic model by a factor of ???2. Significant differences in foreshock rate were found among subsets of earthquakes defined by their focal mechanism and tectonic region, with the rate before thrust events higher and the rate before strike-slip events lower than the worldwide average. Among the thrust events, a large majority, composed of events located in shallow subduction zones, had a high foreshock rate, while a minority, located in continental thrust belts, had a low rate. These differences may explain why previous surveys have found low foreshock rates among thrust events in California (especially southern California), while the worldwide observations suggests the opposite: California, lacking an active subduction zone in most of its territory, and including a region of mountain-building thrusts in the south, reflects the low rate apparently typical for continental thrusts, while the worldwide observations, dominated by shallow subduction zone events, are foreshock-rich. If this is so, then the California generic model may significantly underestimate the conditional probability for a very large (M ??? 8) earthquake following a potential (M ??? 7) foreshock in Cascadia. The magnitude differences among the identified foreshock-mainshock pairs in the Harvard catalog are consistent with a uniform distribution over the range of observation.