The original well-mixed tissue model for the arterial spin tagging techniques is extended to a two-compartment model of restricted water exchange between microvascular (blood) and extravascular (tissue) space in the parenchyma. The microvascular compartment consists of arterioles, capillaries, and venules, with the blood/tissue water exchange taking place in the capillaries. It is shown that, in the case of limited water exchange, the individual FAIR (Flow-sensitive Alternating Inversion Recovery) signal intensities of the two compartments are comparable in magnitude, but are not overlapped in time. It is shown that when the limited water exchange is assumed to be fast, flows quantified from the signal-intensity difference are underestimated, an effect that becomes more significant for larger flows and higher magnetic field strengths. Experimental results on cat brain at 4.7 T comparing flow data from the FAIR signal-intensity difference with those from microspheres over a cerebral blood flow range from 15 to 150 mL 100 g(-1) min(-1) confirm these theoretic predictions. FAIR flow values with correction for restricted exchange, however, correlate well with the radioactive microsphere flow values. The limitations of the approach in terms of choice of the intercompartmental exchange rates are discussed.