The precise three-dimensional (3-D) segmentation of airway from CT image is essential for the image-guided therapy, which helps avoid a serious airway injury. We proposed a new segmentation algorithm for the calculation. Firstly, region growing method was employed to segment the main bronchioles (rough segmentation). And then region growing driven by morphology dilation was used to expand the airway region, where centerline of airway tree was extracted. Terminal bronchia (fine segmentation) were segmented along the centerline. Ultimately, rough and fine segmentation results are combined by a logical OR as final airway tree. Quantitative comparisons with 6 sets of manual segmentation results showed that the algorithm could be used to segment up to 9 bronchia, and the average branch sensitivity of 6th was 63.5%, meeting the requirement of airway tree in the image-guided therapy.