Sensory assessors are measuring instruments, and must be accurate, precise
and valid; individual assessors should also show agreement with each other.
Comparisons between assessors can be tackled at two levels of detail: univ
ariate and multivariate. Univariate treatments depend on calculating statis
tics which are displayed in tables or preferably in graphics. Multivariate
methods which allow for differences between assessors include Generalised P
rocrustes Analysis. Only to a very limited extent can the large differences
between assessors be controlled by selection and training. In the absence
of gross impairment, ability to concentrate and vocabulary and identificati
on skills seem to be the best predictors of future performance. The more so
phisticated methods of analysis identified in this paper are necessary to d
o full justice to the data.