Summary: We propose in this paper an active vision approach for performing the 3-D reconstruction of polyhedral scenes. The perceptual visual strategies we have defined allow the system to incrementally build a complete 3-D map of the scene. The exploration algorithm we describe in this paper is based on a prediction/verification scheme managed using decision theory and Bayes nets. It provides a more complete high level description of the scene and partially solves occlusions problems. Experiments carried out on a robotic cell demonstrate the validity of our approach.