DEM generation using LiDAR point clouds has become more and more popular due to the rapid development of Airborne LiDAR system. In order to produce DEM from point clouds, the major process is to filter non-ground points out from the raw data. However each filtering method cannot avoid causing classification errors, how to assess the classification results and to analysis the influences of error classification on DEM quality is an important research. Two methods are usually adopted currently. The first is the error filtering evaluation by the error matrix. The second is RMSE computed with the reference data. In this paper, the terrain relief is taken account and a normalized DEM difference is calculated as the third quality indicator. The ISPRS data is utilized in this paper and the experiments show that the combination of the three methods is suggested for a better quality control.