The willingness of households to pay for prevention against industrial risks can be revealed by real estate markets. By using very rich microdata, we study housing prices in the vicinity of hazardous industries near three important French cities. We show that the impact of hazardous plants on the housing values strongly differs among these three areas, even if the areas all surround chemical and petrochemical industries. We compare the results from both standard parametric and more flexible, semiparametric models of hedonic property. We show that the parametric model might structurally lead to important biases in the estimated value of the impact of hazardous plants on housing values and in the variations of this impact with respect to the distance from the plants.