In the evaillation of mall Y methods of test, t h e t \\'O usual cri tena-precision and aCCl!-racy-are inwmcient. Accuracy is only applica ble wh ere compari sons with a standard can be made. Pr ec i~ion, when interpreted as d egree of r ep roducibiltty, is ]Jot necessarily a measu r e of merit, becallse a me thod may be high ly reproducible merely because it is too crude to detect small variations.To obtain a quantitative measure of merit of test mothods, a no\y concept-se nsitivity-is introdueed. If lV is a measure of some property Q, and lIM its standard de\' iation, the sensitivity of lIf, denoled "' ;II, is defined by the relation "'u=(dM/dQ) / IIM. It follows from this definition that lhe ~cnsit.ivi l v of a test method mayor mal' not be constant fOl' all values of the properly Q. A slatislCcal t est of significance -is derived for the ratio of sensitivities of alternatiye methods of test. Unlike the standa rd deviation and the coefficient of variation, sensitivity is a lllC'aSllre of merit that IS invaria nt with respect lo any functional transformation of the mcasnI"emC' nt , and is therefore independent of the scale in which the measurement is expressed .