Even though densification and social mixing are declared ob-jectives of many nowadays urban planning paradigms, theirsimultaneous implementation isusually questioned by differentactors and is not frequent in practice. In a market economy,understanding potential demand for this class of development,from different types of households, is essential to define publicpolicies oriented to achieve both compact development (CD)and social mixture. To understand the preferences of house-holds and potential demand, we implement a location choicemodel based on a bid–rent framework and spatial latent classes(LC), using census data and location attributes. By using spatialLC, we do not impose exogenous definitions of which zonesare perceived as CD or suburban, rendering a robust method toidentify variation in preferences. We apply the model toSantiago de Chile, where social mixing in dense and well‐located areas is being intensely discussed. We find strong dif-ferences in households’ valuation of attributes between spatialclasses. Results show that social mixing is more difficult indense, well‐connected areas than in suburban areas becausehigher‐income households are more sensitive to the socio-economic context of the location in compact areas. Besidesshowing evidence on household preferences and their im-plications for social‐mixing policies, this paper also provides aproof of concept for the use of spatial LC (proposed in previouswork by the authors), showing this is a robust methodologyallowing to generate behavior‐based classifications for urban areas. The paper also contributes methodologically, by derivingthe elasticity formulation for bid‐auction location choicemodels, which allows quantifying the importance of locationattributes in location probability.