A general approach to comparisons in the presence of temporal drift is
presented and comparison sequences are discussed for linear, quadrati
c and cubic drift. Circular comparison sequences of the form (ab..)n a
nd (ab..)(n)a are proposed (where a, b,.. are different items, such as
masses, and n is a repeat factor) with the best performance achieved
when three to five items are involved in the comparison. A matrix leas
t squares model is used to obtain information about the drift, the ite
m differences and their uncertainties. The emphasis of the paper is on
mass comparisons, where the dominant drift is in the balance zero, bu
t the approach is applicable to comparisons generally and its applicat
ion to gain drift is also presented.