The present study describes the different approaches to process Geochemical
Reference Material (GRM) databases and the application of an scheme involv
ing fourteen statistical tests to establish concentration data in live GRMs
from the U.S. Geological Survey (basalt BHVO-1, andesite AGV-1, rhyolite R
GM-1, and diabases W-1 and W-2). These tests include deviation/spread, Grub
bs-type, Dixon-type, and high-order moment statistics. For about 50% of the
cases, final mean values are more reliable (characterized by smaller stand
ard deviations) than those obtained earlier using the inadequate "two stand
ard deviation" method. Although about 53% of present mean data are identica
l (within 1 %) to those reported previously, a disagreement >1 to 5% is obs
erved in 32% of the cases. The remaining mean values show larger difference
s (20-50%) compared to the literature data. The present procedure of outlie
r detection and elimination is therefore recommended in the study of GRMs,
instead of the probably erroneous "two standard deviation" method. Use of g
eochemical criteria indicates that the new mean values in basalt BHVO-1 mig
ht be closer to the "true" concentrations.