To accurately describe the circular structures of industrial CT images, a measurement algorithm based on multi-scale geometric analysis tool, named Arclet, was presented. Firstly, a fast mono-scale Arclet numeric transform algorithm was designed through analysis of the spatial relationship between mono-scale Arclet basis functions. Then, the candidate circular features at mono-scale were extracted based on the mono-scale numeric transform transform. By considering the spatial relationship between Arclet basis functions at adjacent scales and the multi-scale tree structure, the candidate circular features between adjacent scales were chosen from a root to leafs. Finally, the surplus circular feature of each scale was fused to obtain the extracted results, and to calculate the related parameters such as radius based on the extracted results. The circle measurement for an actual industrial CT image was performed and the measurement results show that the absolute value of maximum radius absolute error is less than 0.1 mm and that of maximum radius relative error is less than 5%. Even if adding the Gauss noise of different intensities into the original industrial CT image, the measurement results are still can meet the requirements. The measurement algorithm based on fast Arclet transform has better ability to inhibit the noise interference and meet the requirements of industrial CT images for circular structure description.