PSHA models typically use instrumental seismic data, but seismic networks are too young to record many large (and thus infrequent) earthquakes. Nevertheless, PSHA source models should be consistent with the instrumental record. Comparing instrumentally observed earthquakes with fault models can be complicated because large earthquakes may not occur precisely on prescribed faults and they may involve multiple faults. Some tests amenable to probabilistic evaluation include participation (occurrence of rupture on specified fault sections) and breaching (rupture crossing specified boundaries along faults). The relevant sections and boundaries may but need not correspond to “segments” as assumed in characteristic earthquake models.   The Working Group on California Earthquake Probabilities has published increasingly sophisticated earthquake rupture models since 1988, each specifying fault sections with rates of participation, breaching, or both. Here I compare some of the earlier models with instrumentally observed earthquakes both before and after the model start date. I only count earthquakes after 1933, when seismic networks became adequate. Some discrepancies occur even in retrospective comparisons, because the models were not constrained to fit only those seismic data. The discrepancies don’t invalidate any of the models, but they deserve consideration because the models all share fault geometry and assumptions relating slip rate to earthquake rate.