There are situations in the real estate market in which a large number of properties have to be valued at the same time. In such cases it is advisable to use mass valuation methods. These methods involve estimating the value of a property on the basis of the values of the attributes defining it. The aim of the paper is to calibrate the influence of attributes on unit values of properties in mass appraisal in order to minimise the valuation error. The research was conducted for 318 residential properties located in Szczecin. The Szczecin Algorithm of Real Estate Mass Appraisal was used along with the econometric, statistical and expert approaches. The econometric approach is based on the ridge regression model, the statistical approach on the partial Kendall τ correlation coefficients, and the expert approach on the AHP method. The quadratic programming was co-employed with the statistical and expert approaches in order to minimise the mean square error (MSE) of the valuations. The econometric and statistical approaches with the minimisation of the MSE generated best results. The least accurate results were obtained by means of the statistical and expert approaches without the minimisation of the MSE. However, even though the optimisation of the MSE improves the quality of valuations, it also narrows down their volatility, which might make the valuation of properties from the outside of a given database more problematic.