In simulations of plasmas using particle-in-cell methods, it is often advantageous to resample the particle distribution function to increase simulation accuracy, to reduce the computing cost, or to avoid numerical instabilities. An algorithm for down-sampling the particles to a smaller number is proposed here. To minimize noise introduced by the down-sampling, the algorithm is designed to preserve important properties of the particle distribution, particularly low-order moments and local features of the distribution. Tests demonstrate that the proposed algorithm yields a more accurate representation of distributions than simple random down-sampling. The algorithmic approach can also be applied to other resampling problems.