This paper presents two original automatic algorithms for the point segmentation and vectorization of a laser scanning survey of urban buildings. The segmentation algorithm is composed of two sequential phases: in the first one a nonparametric regression model is applied to identify the initial clusters of geometrically and topologically homogeneous points. To each cluster, a parametric model is successively applied to obtain a high reliable roof plane segmentation. The vectorization algorithm is able to distinguish and correctly reconstruct internal and border lines of a roof. By applying the two algorithms to high frequency laser survey data, a reliable vectorial solid modelling of buildings has been automatically obtained.