The ability to track motion from cine phase-contrast (PC) magnetic resonance (MR) velocity measurements was investigated using an in vitro model. A computer-controlled deformable phantom was used for the characterization of the accuracy and precision of the forward-backward and the compensated Fourier integration techniques. Trajectory accuracy is limited by temporal resolution when the forward-backward technique is used. With this technique the extent of the calculated trajectories is underestimated by an amount related to the motion period and the sequence repetition time, because of the band-limiting caused in the cine interpolation step. When the compensated Fourier integration technique is used, trajectory accuracy is independent of temporal resolution and is better than 1 mm for excursions of less than 15 mm, which are comparable to those observed in the myocardium. Measurement precision is dominated by the artifact level in the phase-contrast images. If no artifacts are present precision is limited by the inherent signal-to-noise ratio of the images. In the presence of artifacts, similar in magnitude to those observed in vivo, the reproducibility of tracking a 2.2 x 2.2 mm2 region of interest is better than 0.5 mm. When the Fourier integration technique is used, the improved accuracy is accompanied by a reduction in precision. We verified that tracking three-dimensional (3D) motion from velocity measurements of a single slice can lead to underestimations of the trajectory if there is a through-plane component of the motion that is not truly represented by the measured velocities. This underestimation can be overcome if volumetric cine phase-contrast velocity data are acquired and full three-dimensional analysis is performed.