In late August 2005, a swarm of more than a thousand earthquakes between magnitudes 1 and 5.1 occurred at the Obsidian Buttes, near the southern San Andreas Fault. This swarm provides the best opportunity to date to assess the mechanisms driving seismic swarms along transform plate boundaries. The recorded seismicity can only explain 20% of the geodetically observed deformation, implying that shallow, aseismic fault slip was the primary process driving the Obsidian Buttes swarm. Models of earthquake triggering by aseismic creep can explain both the time history of seismic activity associated with the 2005 swarm and the ?1 km/h migration velocity exhibited by this and several other Salton Trough earthquake swarms. A combination of earthquake triggering models and denser geodetic data should enable significant improvements in time-dependent forecasts of seismic hazard in the key days to hours before significant earthquakes in the Salton Trough.