Using the marginal value of characteristics, we show how to construct bounds on the exact hedonic price index. When prices are above marginal costs then our bounds still apply, but the value of characteristics cannot be measured so easily from a hedonic regression. Since the price--cost markups are an omitted variable, they will bias the coefficients obtained. For a special class of utility functions, we argue that a linear regression will still provide a measure of the marginal value of characteristics, but a log-linear regression will overstate these values.