Abstract The aim of this paper is to model painting prices at auction. The novel aspects of our contribution are as follows: first, the set of regressors used as explanatory variables in the hedonic regression is wider than those previously employed in the literature. Second, we consider the selection bias arising from the possibility of unsold items. Finally, a model including pre sale evaluations by experts is also estimated which allows us to evaluate their information content. To do so, we use the Heckit model exploiting a unique dataset of 2817 Italian Contemporary Art painting transactions which took place at auction worldwide between 1990 and 2006. Our results suggest that auction prices depend upon four sets of regressors (artist identity, physical, artistic and sale characteristics of the painting); moreover, auction house, marketplace and year of sale seem to be crucial in getting artworks sold. Pre sale estimates seem to be a good predictor of painting prices but the hypothesis of their sufficiency is rejected and problems regarding the economic interpretation of the results arise.