Adiabatic fast passage (AFP) is used in noninvasive quantitative perfusion experiments to invert (or label) arterial spins. Continuous arterial spin labeling (CASL) experiments conducted in vivo often assume the inversion efficiency based on the labeling field and steady flow conditions, without direct verification. In practice, the labeling field used in CASL is often amplitude-and duty cycle-limited due to hardware or specific absorption rate constraints. In this study, the effects of the labeling field amplitude and duty cycle, and flow dynamics on the inversion efficiency of AFP were examined under steady flow conditions in a saline flow phantom. The experimental results were in general agreement with models based on Zhernovoi's theory except at high labeling field amplitudes, when the spin inversion times are at least half of the duration of the labeling pulse. The nonlinear relation observed between the inversion efficiency and the labeling duty cycle implies that the practice of linear derating the inversion efficiency with the labeling duty cycle may be prone to significant error. A secondary finding was that the Key words: arterial spin labeling; adiabatic fast passage; inversion efficiency; steady flow; labeling duty cycle During the last decade there has been a significant amount of development in noninvasive quantitative perfusion MRI using arterial spin labeling (ASL), with applications including the human brain (1-5) and kidney (6 -8). Continuous ASL (CASL) employs adiabatic fast passage (AFP) (9) to invert arterial spins (i.e., in blood) prior to their arrival at the tissue region of interest (ROI). The perfusion signal is directly proportional to the inversion (or labeling) efficiency of AFP within a finite inversion region and is attenuated by T 1 relaxation that occurs in transit from the inversion region to the imaging region (5,6,10). It is desirable to maximize the inversion efficiency (and, where possible, minimize T 1 relaxation in transit) since the perfusion signal is small in humans: approximately 1-2% of the image signal for gray matter (11) and 4 -6% in the renal cortex (6,7).Measurement of inversion efficiency in vivo is difficult since it may vary during the cardiac cycle (12) and the ability to synchronize the measurement with the spin labeling (or tagging) is challenging (6). Inversion efficiencies have been measured directly in the rat carotids (13) and rat descending aorta (14), human descending aorta (6) using gated acquisitions, and indirectly in the human brain (4,5,11,15,16) by varying the labeling field.However, the labeling fields and inversion efficiencies are not typically measured in every subject or study (3,17), since such measurements significantly extend the exam time and are prone to large errors. Instead, the inversion efficiency is assumed based on earlier experimental measurements or simulations. Sources of error include the assumptions of the spin hemodynamics (e.g., pulsatile vs. steady flow), relaxation times, labeling field amplitude, and the cycling of the ...