Shirley (1977, Biometrics 33, 386-389) proposed a nonparametric version of Williams' test for determining the lowest dose level at which there is evidence of a toxic effect. A modification of Shirley's procedure is now proposed, which improves its power.