The concentrations of numerous constituents must be evaluated in rock samples if they are to be used as reference materials. Because of varying degrees of specialized skills in earth‐science laboratories, the collaboration of many institutions is required in order to establish reliable values. Analytical data reported in such operations are highly variable. The number of constituents reported per sample varies from one laboratory to another. The number of values reported varies from one constituent to another. The reported values for each constituent are generally incoherent, often displaying skewed distributions. The problem is then how to deduce the most probably “true’ concentration for each constituent. Various computation schemes, based on medians, means and related measures, are discussed. A new scheme, involving a combination of modes, is pro