Consider multivariate linear calibration of a single standard. We show that a selection of the q' most informative of the q responses gives a finite mean squared error for the generalised least squares prediction of an unknown standard provided 1 [less-than-or-equals, slant] q'...