Improvements in high-performance particle-based computer simulations demand a commensurate increase in the performance of analysis codes. We address this need with freud, a Python library containing TBB-parallelized C++ routines for various analysis techniques. All inputs and outputs are NumPy arrays, enabling integration with the scientific Python ecosystem. In addition to standard analyses used in molecular simulation, freud implements novel methods for tasks such as characterizing local order. Used in conjunction with simulation software like HOOMD-blue, freud can dynamically analyze simulations on-the-fly, allowing users to study phase transition kinetics and other complex phenomena. Vyas Ramasubramani, Bradley Dice, Eric Harper, Matthew Spellings, Joshua Anderson and Sharon C. Glotzer