One of the more disappointing failures of the G1-W1 program was that it did not lead to a serious attempt to identify the various sources of significant error in "classical" silicate analysis and evaluate the contribution each makes to the error variance. It is difficult to appreciate now the combination of apathy and antipathy with which most silicate chemists and petrologists alike then viewed the whole subject of statistical analysis. One cannot make much of a case for intelligent experimental design if one cannot convince oneself that the second moment is a potentially meaningful sample statistic, but in our group--the avant-garde of its day--there was a powerful sentiment to the effect that even the first moment should not be computed from data of the sort we were collecting! Considering the situation as a whole, the G 1-W 1 program was immensely informa-