This thesis aims at improving our understanding of the real estate prices formation and of the mechanisms that enable the offer and the demand to meet in what may be a thin market. It participates in the growing literature of the housing market microstructure, in particular the less studied demand side.This thesis differs from the literature by the data it mobilizes, the observations being made through data collected on a digital platform, specialized in real estate price information and estimation.It consists of three empirical studies:The first study consists of an empirical application of matching models on the real estate market. More precisely, thanks to the estimates made on the platform, we construct proxies of the number of buyers and sellers active in the markets of 40 large urban areas of France between 2013 and 2017. Crossing these indicators with the number of sales recorded by the tax authorities in these markets, we perform, to our knowledge, the first estimate of a matching function of the housing market. The main result of this study is to show that contrary to the postulate of the theoretical literature, the returns to scale of this function are not constant, but decreasing.The second study uses the successive estimates made by a user who has first declared himself as a buyer and then as the owner of an apartment. The dataset we have constituted allows an empirical study of the buyer's problem which takes into account buyer’s search itself, and not just the constraints linked to its initial conditions. With this overview of the chronology of visits, the properties they relate to and estimates of the value of those properties, we analyze how the history of the acquisition influences the price paid for a given apartment. It appears that when faced with the uncertainty of apartment values, buyers adjust their internal reference price based on their recent experience. In particular, we measure that a buyer who visits apartments more expensive (resp. cheaper) than the one finally acquired pays more (resp. less), all other things being equal.The last study of this thesis examines the respective capacities of individuals on both sides of an ongoing transaction to predict the final price of this transaction. To do this, we compare the prices of sales recorded in the notarial databases with the estimates made ,in the year preceding the date of the transaction, by the users themselves, after having discovered the result of the platform estimation tools. Unlike sellers who, according to the established result, overestimate the value of their property, buyers do not show any positive or negative bias. Their opinion of the fair price is also less influenced by the result of the platform estimation tool than that of sellers, even if its influence on the latter tends to diminish as they move forward in the sales process.