[en] To avoid underestimating the uncertainty of the evaluated values for sets of consistent data the following rule is proposed: if the smallest of the input measurement uncertainties (σ_min) is more than the uncertainty obtained from statistical data processing, the σ_min should be used as a final uncertainty of the evaluated value. This rule is justified by the fact that almost any measurement is indirect and the total uncertainty of any precise measurement includes mainly the systematic error of the measurement method. Exceptions can be only for measured data obtained by essentially different methods (for example, half life measurements by calorimetry and specific activity determination)