We investigate the structure of demand by focusing on the distribution of prices within narrowly-defined classes of goods. We observe considerable heterogeneity-products that are functionally similar but presumably of different 'quality' may sell at very different prices. We analyze distribution of prices for bottles of wine, used cars, houses in London and week-long holidays in Majorca, and observe in each case that the the resulting distribution is more skewed than the lognormal but less skewed than a Pareto distribution. We then present a theoretical model whereby products can distinguish themselves along multiple hedonic dimensions of 'performance', with these product attributes being random variables subject to multiplicative interactions. Variations of this model can reproduce a lognormal price distribution and a Pareto distribution as lower and upper bound benchmarks (respectively). © The Author(s) 2009.