Pore size distribution (PSD), which is defined as the percentage of the pores of each size to the total volume of the pores, is a key parameter to characterize the microstructure of porous media. Based on the pore geometry definition of three-dimensional (3D) maximal ball and two-dimensional (2D) maximal disk in discrete space, an improved algorithm is proposed to obtain PSD from pore space images of porous media. To validate the accuracy of our algorithm, a synthetic porous medium with a known PSD is generated by computer simulation. In addition, other 2D and 3D images of various types of samples, including sandstones, carbonates, and man-made materials, were also used to obtain PSD. The PSDs were compared quantitatively with that obtained by using the existing state of the art algorithm. The results show that our algorithm is more accurate in characterizing the pore structure and more consistent with the direct visual identification. In addition, the efficiency of our algorithm was also evaluated by calculating the PSD of large-scale 3D images, the results of which indicate that all calculations can be completed in a relatively short time.