To accurately extract the features from massive laser scanning point cloud data, a feature extraction method of laser scanning point cloud based on morphological gradients is proposed. The method first generates the digital elevation model of massive laser scanning point cloud, and then obtains the gradient of each laser footprint by mathematical morphology theory defined by mathematical morphology. The mean value of gradient local nearest points is used as local adaptive threshold. The point cloud data is divided. The characteristic part and the flat part are generated. The random sample consensus method is used to fit the plane from flat part and circles from characteristic part, then the characteristic information such as the height of the step and the radius of the hole is obtained. The experimental results show that the proposed method can effectively extract the features of massive point cloud data. The maximum error of the circles' radius is not more than 0.05 mm, and the minimum error of the step height is not more than 0.1 mm.