Model risk is the uncertainty in model-derived point estimates of unobserved risks. This paper studies model risk in the setting of residential physical risk assessment by analyzing the outputs of two prominent risk modelers. We document five facts about model risk in physical risk models, defined as the absolute difference in the two modelers' estimates of average annual loss (AAL) for the same property and set of perils (flood, hurricane wind, and wildfire). (i) For the average single-family residence, model risk is expected to cost $322 per year, which is approximately equal to the mean of each modeler's AAL estimates. (ii) Aggregating the risk estimates up to the tract or county level does not substantially reduce the disagreement. (iii) Model risk is disproportionately borne by economically vulnerable households. (iv) There is little association between model risk and insurance premiums. (v) Net migration rates are higher in places with higher levels of model risk. Taken together, these facts imply that uncertainty about estimates of physical risk is not appropriately priced by economic agents.