Least Absolute Deviation substituting Least Squares Method is introduced to establish the aim function ∑ ni＝1 ｜y(i)－f(xi，β)｜ in the estimation of position, scale and shape parameters in the extreme value distribution. The extreme value in the stochastic process is the maximum or minimum value on a given time trial, so the extreme method emphasizes particularly the tail of a distribution instead of the whole, which makes sense to forecast the disasters, reduce or avoid losses and so on. As one of the most important step, parameter-estimating attracts more and more consideration. Compared with the classical estimation, the new method using LAD and nonlinear regression model improves in many aspects. It has been found that LAD can enhance not only the accuracy of the forecast but also the calculation stability.