During recent years, the contingent valuation (CV) method has been wid
ely used to value non-marketed goods and services. We present the resu
lts of a CV study of the value of road safety. We find that stated pre
ferences for road safety exhibit considerable imprecision, appear subj
ect to various systematic biases, and are insensitive to variations in
the quantity and quality of the safety improvements concerned. One br
oad implication of these findings may be that, for an important class
of goods (of which safety is one example), standard assumptions about
the structure of people's preferences may be much too strong. A more s
pecific implication, concerning the design and conduct of CV surveys,
is that the NOAA Panel's widely cited blueprint for 'good CV practice'
may rely far too heavily on assumptions about the precision and sensi
tivity of people's preferences.