The paper presents a two-stage method of real estate estimation, basing on the probabilistic model of the variability of real estates prices in relation to their attributes. At the first analysis stage, on the basis of the market information being a base for estimation, a linear model and a multiplicative model of variability of real estate prices in relation to their attributes will be estimated. The verification of statistical hypotheses concer-ning the differences between the values of the estimated model parameters will be the basis for evaluating their significance and it will permit to choose a more reliable model. On the basis of model parameters and of the attributes of an estimated real estate, we will determine its market value with a full analysis of variance. A real estate value determined in such a way, being a prediction of estimated model, will be corrected with a random correction, which is going to be determined at the second stage of the analysis. At the second stage, the statistical analysis will include random deviations, determined at the first stage, relating to the real estate unit prices, together with their covariance matrix. For this analysis, a procedure based on quality methods will be evolved with the aim of determining the degree of similarity between an estimated real estate and a real estate in the database. As result of the quality analysis, a subgroup of real estates showing the highest degree of similarity to the estimated real estate will be established. Random deviations for real estate unit prices of the established subgroup and their covariance matrix, determined at the first stage, are the basis for determining the random correction for the estimated real estate. This random correction being added to the real estate market value obtained from the model prediction determined at the first stage, we get the most probable market value of the estimated real estate.

3-13