We develop an extension to the method of Boatwright and Choy <1986> for determining the radiated seismic energy Es that accounts for factors that bias the estimate. We apply our technique to 204 events worldwide during the period 1992--1999 and find that the apparent stress is on average largest for strike-slip events (0.70 MPa), while for reverse and normal events it is significantly smaller (0.15 and 0.25 MPa, respectively). These results support the mechanism dependence of Es reported by Choy and Boatwright <1995>, although we find that once likely sources of bias are accounted for, the mechanism dependence is not as strong as found previously. The source of the mechanism dependence is unclear, but one possibility is that it reflects a mechanism-dependent difference in the stress drop. This hypothesis is suggested by the scaling of slip with width in large strike-slip earthquakes and makes two predictions, which could be used to test it. The first is that the discrepancy should disappear for the very largest dip-slip earthquakes as the length of the fault greatly exceeds the downdip extent. The second is that the discrepancy ought to disappear for smaller earthquakes. The first cannot yet be tested due to a lack of recent, very large dip-slip earthquakes. The second is supported by the lack of mechanism dependence to Es for smaller earthquakes. An alternative hypothesis is that the apparent mechanism dependence could result if faults are opaque during rupture, blocking seismic radiation across them <Brune, 1996>. This could cause radiated seismic energy to be trapped preferentially in the crust near the source volume for dipping faults. There remains, however, a large discrepancy between estimates of Es obtained from teleseismic versus regional data. This discrepancy indicates a problem with teleseismic and/or regional estimates of the seismic energy and must be resolved before a definite conclusion can be drawn. ¿ 2001 American Geophysical Union