The airborne laser scanning (ALS) data is becoming a standard approach for generating digital elevation model (DEM) in recent years. To generate the DEM, the ground points in the point clouds need to be classified firstly. Traditionally, the points provided by the multi-return LiDAR system only provide the three-dimension coordinates of points and the classification approaches normally can mainly utilize the geometric conditions to separate the ground points from the non-ground points. If the terrain relief is varied quickly, the geometry-based classifier could not be satisfied for ground filtering purpose. In recent years, the commercial waveform LiDAR system has advanced developed. The waveforms can preserve more information of targets than the traditional multi-return LiDAR data. Since waveform is the interactions between targets and laser signals, this interaction can be considered as the convolution between the transmitted laser pulse and the backscatter cross-section function of the targets. For this reason, more radiation and geometric characteristics of targets can be extracted. As the observations of real waveform data, we found out that the width of trees is larger than the ground. For this reason, we utilized the backscattered cross-section of targets and the echo width extracted from waveform data as new properties of points for improving the classification accuracy. Firstly some training data is chosen for designing the classifier to separate the non-ground points from the point clouds. We aimed at the analysis of the performance in classifying ground points by using the extracted waveform features. Our preliminary results show the potential of the waveform features for ground classification purpose in a forest area and also reveal thatthe characteristics of waveform is useful in classification.