Most probabilistic seismic hazard studies assume that future earthquakes will occur in clearly defined, seismically homogeneous areal source zones. This assumption means that modeled seismic rates change abruptly at source zone boundaries, with the result that predicted acceleration levels may differ considerably at sites a few kilometers apart near a boundary (e.g., changes of 50 to 80 per cent or more at sites 20 km apart). However, homogeneous source zones are at best an approximation, and the areal extent of such zones is usually not well determined. I propose modifying the model so that the seismicity originally associated with each point in a source zone instead be regarded as normally distributed with standard deviation σ about that point; this permits seismicity to vary smoothly across source zone boundaries. When this variability is modeled, if σ is 25 per cent or less of the width of the source zone, acceleration levels calculated at sites at the center of the zone and at sites on the boundary are about the same as those calculated when no location variability (σ = 0) is assumed; however, calculated levels decrease at sites between the center and the boundary, and increase at sites outside the boundary. The changes tend to become larger at the higher accelerations corresponding to longer exposure times.
When earthquakes are modeled as ruptures along well-defined linear faults, predicted accelerations also increase significantly as the site location approaches the fault. However, in many cases, neither the location of a fault nor the definitions of its end points can be precisely stated. Acceleration as a function of site location will vary more smoothly if possible fault locations are regarded as either normally or uniformly distributed.