The accuracy of magnetic resonance phase contrast volume flow measurements in small blood vessels is expected to be smaller than in large vessels, because of partial volume effects at the vessel boundary. Accuracy was validated in the dog femoral artery, diameter 3.5 +/- 0.7 mm, using an ultrasonic transit-time flowmeter (TT). The number of pixels per vessel diameter (ND) ranged from 1.6 to 4.8. The vessel cross-section was determined using a threshold in the magnitude image. Between the two methods the correlation coefficient was 0.95 (range 10-200 ml/min). The proportional difference (PD), (QTT-QMR)/1/2(QTT+QMR), was 0.8%, showing no systematic difference between the methods. The PDs standard deviation was 27%, and 19% for flow rates above 30 ml/min. Only a significant decrease of the PDs variance was found at the highest ND values, suggesting other sources of error than partial volume effects. It is concluded that with an ND value of about 3, accurate blood volume flow rates can be determined.