We have examined the historical seismicity of the Mexicali-Imperial Valley. Since about 1850, the largest observed earthquake in the area has been the magnitude (Mw) 7.1 December 1934 event. The magnitude distribution for large plate-boundary rupturing earthquakes (M > 5.8) is not well fit by the Gutenberg-Richter type (b value) model. Models which better fit the observed magnitude distribution are derived from a lognormal distribution of stress drops on faults of constant size and from an exponential distribution of rupture lengths, as might be expected from randomly distributed rupture-stopping asperities along the length of the fault. This may be among the best data in the world for testing earthquake recurrence models (e.g., time predictability, slip predictability) on a strike-slip fault, but the uncertainties are still too great to permit a unique interpretation.