It is reasonable to expect that the number of aftershocks following a large earthquake should reflect the degree of heterogeneity on and near the fault plane. To investigate this question, we plot N, the number of aftershocks with mb ≧ 5.0 occurring within the next 30 days after large, subduction-zone, thrust-faulting earthquakes (Mw ≧ 7.0), versus the magnitude (Mw) of the main shock. Only earthquakes that occurred after the installation of the World-Wide Standard Seismographic Network stations are considered. Because the detection threshold of aftershocks in many regions is greater than mb = 5.0, N is obtained by fitting the data to Gutenberg and Richter's relation (log N = a − b · mb). For the same Mw, N varies by a factor of about 40 in different regions. N is less than expected in the eastern Pacific (including Alaska) and is greater than expected in the western Pacific. Relative lack and excess of N correlates well with regions of supposedly strong and weak coupling, respectively. While small and large N also show some correlation with the expansion rate of aftershock area with time, the variation in N is much larger than would be expected from the mere increase of aftershock areas. In some cases, it appears that the number of aftershocks, N, for earthquakes of similar Mw, may also reflect the complexity and enrichment of the P-wave amplitude spectra from 1 to 10 sec. For example, the Valparaiso, Chile, earthquake of 1985 (Mw = 8.0) generated N = 46 earthquakes, whereas for the Michoacán, Mexico, event of 1985 (Mw = 8.1), N is only equal to 5. Furthermore, for the Valparaiso earthquake, P waveforms are much more complex, and the spectra are more energetic at 1 to 10 sec as compared to the 1985 Michoacán event.