Recently, several implementations of arterial spin labeling (ASL) techniques have been developed for producing MRI images sensitive to local tissue perfusion. For quantitation of perfusion, both pulsed and continuous labeling methods potentially suffer from a number of systematic errors. In this study, a general kinetic model for the ASL signal is described that can be used to assess these errors. With appropriate assumptions, the general model reduces to models that have been used previously to analyze ASL data, but the general model also provides a way to analyze the errors that result if these assumptions are not accurate. The model was used for an initial assessment of systematic errors due to the effects of variable transit delays from the tagging band to the imaging voxel, the effects of capillary/tissue exchange of water on the relaxation of the tag, and the effects of incomplete water extraction. In preliminary experiments with a human subject, the model provided a good description of pulsed ASL data during a simple sensorimotor activation task.