The principal procedures for the statistical analysis of gold assay and other trace-element data depend on applying either the normal frequency distribution or the lognormal frequency distribution. The consequences of applying both methods are examined through reference to example data from United States and South African gold mines. It is concluded that for estimation of average grades, except for certain data with a very high coefficient of variation (ratio of standard deviation to mean), use of the lognormal distribution is ordinarily undesirable because bias may be introduced and little efficiency can be gained. However, if biased average grades can be tolerated, as in the examination of data from surveys in exploration geochemistry, use of the lognormal distribution may be desirable. Various statistical analyses are detailed, and the tables presented to implement the analyses are more complete than those previously published.