Our recent adaptation to PET of the method of Fitchard et al. [1], [2], [3] for rigid body registration of CT sinograms enables motion between two temporal frames of PET data to be estimated and corrected prior to reconstruction. This avoids both the computation required by multiple reconstructions and the need to make choices regarding reconstruction methods that influence the images produced, and potentially change the estimated motion. Using realistic, simulated data with known ground truth, we report an initial investigation into the performance of the method as the number of counts and the accuracy with which the scan is divided into frames, corresponding to different positions, varies.