Several statistical restrictions, which are critically important for correct use of different Big Data procedures in metallurgy for attestation and management of quality of metal products are evaluated. Representative production control data stores for steel manufacturing technologies are used as the research object. This research covered wide grade and dimension range of steels: large forgings of heat treatable 38KhN3MFA-Sh steel, rolled products of 40KhMFA steel, sheet 17G1S-U, 09G2S and 15KhSND steels. Possible scale of variety of values distribution both for managing parameters and characteristics of strength, plasticity and toughness is shown using the coefficients of asymmetry and kurtosis. These characteristics are varied within the technological tolerance range. Accompanying risk during metal quality prediction and management, e.g. using the methods of parametric statistics, was evaluated for the case when this circumstance was not taken into account. The features of influence of a sample list volume on the results of statistical processing of large production control data stores and metallurgical product are revealed. It is shown how absence of common space of parameters restricts possibilities of classic statistics in metallurgy, makes non-effective the management "by disturbance" principle. In this connection, possibilities of non-parametric statistics, presented by Kolmogorov -Smirnov criterion, which is not depended on distribution of collection of analyzed sample lists, are evaluated. To provide objective selection of the areas with dominating type of relationship, it is necessary to take into account possibility of existence of different evolution scenarios for structure and defects along the technological chain (technological heredity) within the framework of rather wide tolerance range, as well as features of their appearance. Difference in the evolution mechanisms of structures and defects within the framework of separate technological trajectory is a cause of appearances of developed heterogeneity for nominal single-type structures which have, however, different scales, as well as accompanying quality dispersion (which is often essential). Taking this circumstance into account allows to find out the links in the system "managing parameters -final parameters of metal products", which are not always evident during their search using generally accepted approaches. Development of the complex of rules for online management of metal products quality is possible on this base. Key words: quality management of metal products, retrospective analysis of data bases for production control, Big Data, technological heredity, classic and non-parametric statistics, regression, kind of distribution of parameter values, dominating type of relationship.