Abstract: A modified non-parameter method(M-NPM) for the confidence interval estimation of DEM errors was developed based on NPM.Six different DEMs obtained by an airborne laser scanner were employed to comparatively analyze the accuracy of M-NPM and NPM.The six DEM error populations with three slightly non-normal and three very non-normal distributions were acquired with an across-validation process.Stochastic sampling from error populations allows us to report that when the sampling number is smaller than 40,both NPM and M-NPM are obviously affected by the degree of normality of the population distribution,but the influential degree of M-NPM is smaller than that of NPM.No matter what the population distribution is,and how many the sampling points are,the results of M-NPM are more robust than NPM,which is attributed to the fact that M-NPM presents wider confidence intervals than NPM.