(2)) or the IAUCBA (3) (initial area under the Gd concentration time curve in tumor, blood adjusted) in one or more malignant target lesions upon drug treatment have been recommended as primary measures of drug activity (4). The DCE-MRI results from each lesion are usually summarized by one number, the percentage change of K trans or IAUCBA from baseline upon drug treatment. The DCE-MRI parameters are estimated from the contrast agent (CA) concentration time course in the entire tumor, in a selected tumor subvolume (region of interest), or voxel by voxel (in which case the mean or median parameter estimate across all voxels is extracted to characterize the tumor).K trans is usually estimated by applying the Tofts model, or the "extended Tofts model", having the additional parameter v p (5). The Tofts model was originally proposed to describe gadolinium diethylenetriamine penta-acetic acid (Gd-DTPA) kinetics in multiple sclerosis lesions (6) and in cardiac muscle (7), and several observers have noted that neither this model nor the extended Tofts model is flexible enough to adequately describe low-molecular-weight CA kinetics in most malignant tumors (8-16) (See Fig. 10 in (12)). The lack of fit is most clearly apparent in wholetumor or region-of-interest analyses (Fig. 1) but may also be seen in the more noisy single voxel profiles (11,13,17).The "kinetic heterogeneity" of malignant tumors (8) can be described with more complex compartmental models (8)(9)(10)(11)16,18), but the degree of kinetic complexity may not be the same in all tumors and is never the same in all voxels of a given tumor; it may also change under drug treatment. This raises the question of how to arrive at one or, at most, two summary parameters to characterize CA kinetics in all malignant tumors in a given clinical trial for comparisons among patients and patient visits. A similar problem exists in general pharmacokinetics, where the kinetics of a drug in the circulation may be more or less complex in different patients, on different occasions, at different doses, or with different modes of administration so that compartmental models with varying numbers of compartments are required to describe the kinetics of a drug in different situations (19). This has given rise to the "noncompartmental" modeling approach, which describes linear drug kinetics regardless of the degree of complexity by two global parameters, clearance (CL) and steady-state distribution volume V ss , which are linked to the mean residence time (MRT ) in the system (MRT = V ss /CL) (20). We propose a similar approach for DCE-MRI data from malignant tumors.Noncompartmental kinetic analysis is simple, in principle, in systemic pharmacokinetics (drug kinetics in the circulation). CL is estimated as dose/area under the plasma concentration-time curve and, for drug administration by intravenous bolus, MRT can be estimated as area under the moment curve/area under the plasma concentration-time curve where area under the moment curve is ∞ 0 C p (t) · t · dt (t = time, C p = drug ...