Given a set of representative samples of aggregate source material (e.g. fr
om a stockpile or production line), it is shown that if only percentaged da
ta is reported, the mean composition of the percentaged data is best calcul
ated using a robust estimate of the logratio mean. However, If raw mass dat
a is also available (as will generally be the case), then it is best to fir
st calculate the average of the constituent masses, and then to convert the
se into a percentage. The difference between confidence; prediction and tol
erance intervals is reviewed and it is shown that the repeatability of aggr
egate samples and an effective long-term 'detection limit' for constituents
:can be reliably estimated using duplicates of run-of-the-mill samples; an
upper boundary can be:placed on the detection limit by application of, say,
a two-sided {95%, 99%} tolerance interval. Numerical approximations are gi
ven to enable a variety df normal two-sided tolerance factors to be compute
d.