For the first time, three-dimensional (3-D) high-resolution images of snow were obtained using X-ray absorption tomography. Images with a spatial resolution of 10 μm were taken on four different cylindrical snow samples (9 mm high, 9 mm diameter). About 1000 two-dimensional X-ray absorption images were recorded at angular positions of the object around an axis spanning 180°. An appropriate algorithm was then used for these data to reconstruct a 3-D image. In the case of snow, experimental problems have been solved to prepare the samples and prevent both melting and metamorphism of snow during the experiments. This tomographic method provided 3-D data files from which images of 6003 voxels were extracted Several physical parameters of snow microstructure can be processed from these data. Porosity P and discrete local (3-D) curvature C of the grain/pore interface were computed for the four snow samples. Representative elementary volume (REV, in the sense of porous media) is a relevant index to the significance of the sample size with respect to a given parameter. From each image, the values of P and C are compared for subsamples of different size, as an attempt to assess the REVs for porosity and curvature. Results show that the observed volume of snow is statistically significant to achieve the porosity and the curvature distribution.