Source time functions of 255 moderate to great earthquakes obtained from inversions of teleseismic body waves by Tanioka and Ruff <1997> and coworkers were compared in a systematic way. They were scaled to remove the effect of moment and to allow the direct comparison and averaging of time function shape as well as duration. Time function durations picked by Tanioka and Ruff <1997> are proportional to the cube root of seismic moment if moments from the Harvard centroid moment tensor catalog are used. The average duration of scaled time functions is shorter and the average shape has a more abrupt termination for deeper events than shallower ones, with a distinct change occurring at ~40 km depth. The complexity of the time functions, as quantified by the number of subevents, appears to decrease below ~40 km depth. Furthermore, among events shallower than 40 km, the average duration of scaled time functions is shorter, and their average shape has a more abrupt termination (1) for events with strike-slip focal mechanisms compared to thrust events and (2) for the few thrust events associated with an intraplate setting compared to the majority associated with an interplate (subduction) boundary. In each of these cases, events in more tectonically and seismically active settings have a longer duration and a more gradual termination. This can be interpreted in terms of lower stress drops and/or slower rupture velocities at active plate boundaries, suggesting that fault rheology depends on slip rate and may evolve as total fault slip accumulates. Furthermore, differences in average time function shape and duration associated with different subduction zones suggest that differences exist in the rheology on the plate boundaries at the various subduction zones. ¿ 2001 American Geophysical Union