Arterial spin labeling (ASL) is a potential means of obtaining quantitative images of cerebral blood flow (CBF). However, few validation studies of ASL have been performed in animal models using gold-standard CBF methods. Other methods that use radiolabeled water as a tracer underestimate CBF in high flow states, but this effect has not been evident in ASL studies. In this study the accuracy of ASL measurements of CBF were modeled and experimentally validated, with particular attention paid to high flow rates. The ASL signal as modeled included the contributions from intravascular labeled spins. The modeling demonstrated linearity of the ASL signal with respect to baseline flow, and linearity of ASL signal changes with respect to changes in flow, including high-flow conditions. Validation studies using quantitative autoradiography (QAR) to image flow in a rat model of unilateral cerebral ischemia showed that ASL systematically overestimated CBF by 34%. A similar overestimation was also predicted by modeling. These results indicate that ASL signals are linear with respect to flow (even high flow), but ASL-CBF measurements are systematically overestimated.