Dislocation time histories of models derived from waveforms of seven earthquakes are discussed. In each model, dislocation rise times (the duration of slip for a given point on the fault) are found to be short compared to the overall duration of the earthquake (??? 10%). However, in many crack-like numerical models of dynamic rupture, the slip duration at a given point is comparable to the overall duration of the rupture; i.e. slip at a given point continues until information is received that the rupture has stopped propagating. Alternative explanations for the discrepancy between the short slip durations used to model waveforms and the long slip durations inferred from dynamic crack models are: (1) the dislocation models are unable to resolve the relatively slow parts of earthquake slip and have seriously underestimated the dislocations for these earthquakes; (2) earthquakes are composed of a sequence of small-dimension (short duration) events that are separated by locked regions (barriers); (3) rupture occurs in a narrow self-healing pulse of slip that travels along the fault surface. Evidence is discussed that suggests that slip durations are indeed short and that the self-healing slip-pulse model is the most appropriate explanation. A qualitative model is presented that produces self-healing slip pulses. The key feature of the model is the assumption that friction on the fault surface is inversely related to the local slip velocity. The model has the following features: high static strength of materials (kilobar range), low static stress drops (in the range of tens of bars), and relatively low frictional stress during slip (less than several hundreds of bars). It is suggested that the reason that the average dislocation scales with fault length is because large-amplitude slip pulses are difficult to stop and hence tend to propagate large distances. This model may explain why seismicity and ambient stress are low along fault segments that have experienced large earthquakes. It also qualitatively explains why the recurrence time for large earthquakes may be irregular. ?? 1990.