It is commonly assumed that equilibrium transcytolemmal water exchange in tissue is sufficiently frequent as to be fast on any NMR time scale achievable with an extracellular contrast agent (CR) in vivo. A survey of literature values for cell membrane diffusional permeability coefficients (P) and cell sizes suggests that this should not really be so. To evaluate this issue experimentally, we used a programmed intravenous CR infusion protocol for the rat with several rate plateaus, each of which achieved an increased steady-state concentration of GdDTPA(2-) in the blood plasma. Interleaved rigorous measurements of (1)H(2)O inversion recoveries were made from arterial blood and from a region of homogeneous thigh muscle tissue throughout the CR infusion. We made careful relaxographic analyses for the blood and muscle (1)H(2)O longitudinal relaxation times. The combined data from several animals were evaluated with a two-site model for equilibrium transcytolemmal water exchange. An excellent fitting was achieved, with parameters that agreed very well with the relevant physiological properties available in the literature. The fraction of water in the extracellular space, 0.11, is quite consistent with published values, as well as with reported tissue CR concentrations when one accounts for the restriction of CR to this space. The derived average lifetime for a water molecule in the thigh muscle sarcoplasm, 1.1 +/- 0.4 sec, implies a sarcolemmal P of 13 x 10(-4) cm/sec, which is well within the range of literature values determined in vitro. Moreover, we find that because of the exchange, the (1)H(2)O longitudinal relaxation rate constant exhibits a decided nonlinear dependence on the tissue or thermodynamic (extracellular) concentration of GdDTPA(2-). The muscle system departs the fast-exchange limit at a [CR] value of <100 micromol/L. This has significant implications for the quantitative use of CRs as MRI tracers. Magn Reson Med 42:467-478, 1999. Published 1999 Wiley-Liss, Inc.