Abstract. In diagnosing lung diseases, it is highly desirable to be able to segment the lung into physiological structures, such as the intra-thoracic airway tree and the pulmonary structure. Providing an in-vivo and non-invasive tool for 3D reconstruction of anatomical tree structures such as the bronchial tree from 2D and 3D data acquisitions is a challenging issue for computer vision in medical imaging. Due to the complexity of the tracheobronchial tree, the segmentation task is non trivial. This paper describes a 3D adaptive region growing algorithm incorporating gain calculation for segmenting the primary airway tree using a stack of 2D CT slices. The algorithm uses an entropy-based measure known as information gain as a heuristic for selecting the voxels that are most likely to represent the airway regions.