We present an alternative approach to tomo-PIV that seeks to recover single voxel particles rather than blobs of extended size. Our approach follows the classical MLOS-SMART method with a weight matrix built with the system’s Point Spread Function coefficients. As such an approach requires only a few voxels to explain the image appearance, it yields much more sparse reconstructed volumes. We carefully justify our method for building the weight matrix and show on synthetic PIV images with a large exploration of generating conditions that it always leads to better results than MLOS-SMART.