Heat flow data near the San Andreas Fault (SAF) do not reveal a near‐fault anomaly as expected from frictional heat generation, an observation interpreted to indicate that the fault slips at a depth‐averaged shear stress <20 MPa. The data also contain large unexplained scatter, which has been a separate major issue in the analysis of heat flow within the California Coast Ranges. Here we use numerical models of heat conduction to evaluate the hypothesis that thermal refraction, due to contrasts in thermal conductivity in the subsurface, both produces the observed scatter in heat flow and as a result obscures the thermal signature from frictional heating on a fault that supports large shear stress during slip. Our study focuses on the region around the San Andreas Fault Observatory at Depth (SAFOD) near Parkfield, California. Our results show that surface heat flow is most sensitive to the contrast between Tertiary sediments and basement rocks and to wavelengths of basement topography of ∼10 km. With realistic thermal conductivity contrasts and a reasonable interpretation of this geologic contact, we show that thermal refraction is a plausible explanation for the observed heat flow scatter. However, refraction effects are unable to mask frictional heat generation in a manner consistent with observations. We show that even with large refraction effects, low background heat flow, a regional NW–SE decrease in heat flow, or nonsteady state heat conduction, the data are most consistent with a fault that produces little to no frictional heat.