In recent years a number of physiological models have gained prominence in the analysis of dynamic contrast-enhanced T 1 -weighted MRI data. However, there remains little evidence to support their use in estimating the absolute values of tissue physiological parameters such as perfusion, capillary permeability, and blood volume. In an attempt to address this issue, data were simulated using a distributed pathway model of tracer kinetics, and three published models were fitted to the resultant concentration-time curves. Parameter estimates obtained from these fits were compared with the parameters used for the simulations. The results indicate that the use of commonly accepted models leads to systematic overestimation of the transfer constant, K trans , and potentially large underestimates of the blood plasma volume fraction, V p . In summary, proposals for a practical approach to physiological modeling using MRI data are outlined. The last decade has seen a rapid development in the use of dynamic contrast-enhanced T 1 -weighted MRI in medicine. In tandem with the technological advances that have enabled improved data acquisition, a number of investigators have employed physiological models to facilitate data interpretation. While the use of these models has found numerous applications (e.g., in studies of tumor physiology (1) and myocardial perfusion (2)), and a number of groups have assessed the potential of model parameters as surrogate markers (3,4), little has been published that addresses the direct interpretation of these results. Specifically, how do the estimates obtained using the various models compare with the physiological parameters they purport to measure? This is not a simple question to address since it is often difficult to identify a reliable "gold standard." Many investigators compare their results with those obtained with positron emission tomography (PET). However, PET shares many of the basic models employed in MRI (5). Similarly, data simulation exercises using Monte-Carlo techniques designed to assess accuracy and precision in parameter estimation often utilize the same model to both generate and analyze the data (6,7). In this way, the sensitivity of the estimates to experimental variables, such as noise and sampling frequency, is assessed but little is revealed about the physiological significance of the resultant parameter estimates.A physiological model incorporating multiple parallel pathways and heterogeneous flow was used to simulate data of a realistic nature to which simplified models were fitted. The experiment was designed to assess the accuracy of the models themselves, not the quality of the data to which they are fitted (in terms of noise, sampling frequency, etc.), since data quality is essentially an experimental variable. Furthermore, this study was restricted to those models dealing with a contrast agent that diffuses out of the vascular space (thereby incorporating capillary permeability as a model parameter). Many of the issues associated with the analysis of data from...