Grade point averages (GPAs) are calculated by assigning numbers to letter g
rades and averaging them. Simple examples show that the method cannot consi
stently determine class rank since class rank is sometimes permuted with ar
bitrary change of scale. This permutation is only possible when one student
is somewhere worse and somewhere better than a second. The distinction bet
ween these and other sorts of cases is established by theorems proved in an
appendix. Relativistic attempts to resolve the inconsistency are shown to
be insufficient. The function of GPAs as predictors is briefly discussed.