This paper presents a feature-based approach to piecewise planar modeling of architectural scenes from an oriented image sequence. An improved line detection algorithm is presented, which uses vanishing points to detect lines in the images and reconstruct them in 3D. Subsequently an extension of the 3D Hough transform is presented which allows efficient and robust detection of planes in a set of 3D lines. The detected planes and a dense point-set are used to reconstruct a piecewise planar building model. The presented techniques are demonstrated on a real-world dataset. Keywords 3D building reconstruction, line reconstruction, 3D Hough Transform