Testing for the linearity of the calibration is an important feature i
n the development of an analytical method. The correlation coefficient
is often used in this contest but is open to misinterpretation by the
unwary and is to be discouraged. When data are collected in well desi
gned calibration experiments, tests for non-linearity can be based on
the analysis of the residual variance from a regression into parts owi
ng to 'lack-of-fit' and 'pure error'. The calculations can be executed
by means of a few simple commands in many statistical software packag
es.