Maximum likelihood recursions were used by Wu (1985) to estimate extreme quantiles of a quantal response curve. For certain choices of initial designs, Wu’s method performs well. In many fields of application, there often exist some different initial designs which are known as the up-and-down designs. Based on the existing data set from such a design, the authors propose three sequential empirical Bayesian designs by quickly and efficiently exploiting the information in the testing data and known knowledge. The improvement obtained by using the new procedures for the estimation of extreme quantiles is substantial.