Although x-ray tomography is commonly used to characterize the three-dimensional structure of materials, sometimes this is impractical due either to limited time for data collection (such as in rapidly-evolving systems) or the need to limit the radiation exposure of the sample. In such situations, it is desirable to extract as much information as possible from a more limited data set. In this paper, we describe how to extract the size distribution of non-spherical pores (or, equivalently, particles) from single x-ray phase contrast imaging (XPCI). Because the pores overlap in projection, interpreting the images and extracting quantitative information about the size distribution is non-trivial. In this paper we extend a previously-developed Fourier-based framework for interpreting the speckle pattern of XPCI images from materials with spherical pores to the more challenging case of non-spherical pores. We develop an analytical expression for the XPCI image from a distribution of randomly-oriented ellipsoidal pores, and show that we can use this expression to extract quantitative information about the size distribution from single images. We discuss three approaches to evaluating this expression, corresponding to different assumptions about the nature of the size distribution, and validate our results with simulated XPCI images and experimental data from Berea sandstone.