Earthquake hazard modelers often assume that earthquake ruptures will be confined to individual faults and that earthquakes’ magnitudes will not exceed values related to the mapped fault length. However, earthquakes have been observed to jump from one fault to another, producing a rupture longer than the originating fault. The magnitudes of these earthquakes can exceed the value based on the originating fault length. We estimated the probability of jumping as a function of interfault distance, based on retrospective observations from past earthquakes. A simple model, consistent with the few available observations, suggests that an earthquake on one strike-slip fault may jump to another with a probability that decreases exponentially with separation. The scale distance is about 3 km. The model implies a nonnegligible probability of jumping as much as 10 km. Many California faults are within 10 km of one another; we identified those and wrote a simple program to calculate the interfault distance and thus the jump probabilities. We also formulated a simple way to estimate the magnitude-frequency distribution for a pair of nearby faults as a function of their independent magnitude distributions and their separation distance. Allowing for multifault rupture increases the estimated upper magnitude limits for faults, relative to a model in which faults are independent. If the magnitude-frequency distribution is constrained to be consistent with a fixed moment rate (from fault slip rates, for example) then increasing the upper magnitude limits will reduce the rate of intermediate and small events.