Three-dimensional (3D) shapes of particles are characterized from a two-dimensional (2D) image of the particles that is obtained using TEM. The 3D shape characterization method includes the steps of obtaining a 2D image of a batch of nanoparticles, determining 2D shapes of the nanoparticles from the 2D image, and deriving six distributions, each of which corresponds to a 2D shape and a 3D shape associated with the 2D shape. The first size distribution is derived from the nanoparticles having the 2D triangle shape. The second and third size distributions are derived from the nanoparticles having the 2D tetragon shape. The fourth, fifth and sixth size distributions are derived from the nanoparticles having the 2D round shape. Based on these six size distributions, three size distributions, each of which corresponds to one of three 3D shape classes, are estimated. The size distributions corresponding to the 3D shape classes provide a better log-normal distribution than the size distributions corresponding to the 2D shapes.