Magnetic resonance phase-contrast volume flow rate (VFR) measurement with limited resolution in small vessels is subject to two major sources of error: a) partial volume artifacts, causing systematic overestimation of the VFR, and b) errors related to the selection of vessel pixels [region of interest (ROI)], causing large inter-observer and intra-observer variability. Additionally, limited resolution results in Gibbs-ringing around vessels, which adversely affects VFR determination. In this paper, a semi-automatic model-based method is presented that effectively eliminates errors due to both partial volume effect and Gibbs-ringing and also minimizes errors from variability in the ROI selection. The model assumes a parabolic flow profile and cylindrical vessel geometry, incorporates inflow effects, and takes into account the point-spread function of the acquisition. The method automatically estimates maximum velocity, vessel radius, and VFR. The method is validated in phantoms under various conditions and evaluated in vivo. For small vessels with moderately pulsatile flow, it is demonstrated that accurate VFRs and diameter estimates are obtained, virtually independent of the ROI selection, even in vessels covered by just a few pixels. Compared with conventional VFR analysis, both accuracy and reproducibility improve significantly.