<p>This work deals with assessment of repeatability and reproducibility with the emphasis laid on data sets typical to critical evaluation. Such sets contain data measured by different authors and in different laboratories which means that the differences between them cannot be modelled by a large number of small errors, in other words the data do not have normal distribution and application of traditional statistical tools can yield wrong results. Algorithms of mathematical gnostics are therefore used. Robust regression, both linear and nonlinear, can be applied in situation where measured data are not directly comparable or even cannot be displayed graphically but can be expressed as a dependence of a quantity on temperature, pressure, composition, or another quantity. Error propagation is a complex task which would need its own article, therefore only simple cases are addressed including the effect of rounding errors.</p>